The academic effects of chronic exposure to neighbourhood violence

Abstract

Chinese

This paper estimates the causal effect of repeated exposure to violent crime on test scores in New York City. We use two empirical strategies; value-added models linking student performance on standardised exams to violent crimes on students’ residential block, and a regression discontinuity approach that identifies the acute effect of additional crime exposure within a one-week window. Exposure to violent crime reduces academic performance. Value-added models suggest the average effect is very small (approximately −0.01 standard deviations) but grows with repeated exposure. Regression discontinuity (RD) models also find a larger effect among children previously exposed. The marginal acute effect is as large as −0.04 standard deviations for students with two or more prior exposures. Among these, it is almost one tenth of a standard deviation for Black students. We provide credible causal evidence that repeated exposure to neighbourhood violence harms test scores, and this negative effect increases with exposure.

Keywords

children education neighbourhood effects neighbourhood violence violent crime

Introduction

Exposure to chronic violence is a continuing problem facing children in the United States. While New York City (NYC) has been one of the country’s safest large cities during the past two decades, many of the city’s youth still live in neighbourhoods plagued by violent crime. During 2010, the last year of our data, almost half of NYC’s fourth to eighth graders lived on a block where a homicide or felony assault occurred. Roughly a quarter lived on a block where two or more violent crimes occurred, and over 4000 students lived on blocks with nine or more violent crimes. And violence is growing: 2020 saw a substantial increase in shootings and murders.¹

Abundant evidence shows that children living in violent neighbourhoods suffer negative consequences: they score lower on standardised exams and cognitive assessments and have lower school attendance. Causal evidence of a negative ‘acute’ effect has been estimated by comparing the test scores of children exposed to violent crimes right before an assessment or standardised test to children exposed directly afterwards (Sharkey, 2010; Sharkey et al., 2014). The focus of these studies is a single exposure to violence, but children exposed to violent crime often reside communities where violence is common. These acute effects may underestimate the full impact of neighbourhood crime on achievement as they overlook its role as a chronic stressor. The literature on neighbourhood disadvantage documents the negative consequences of chronic stress on myriad outcomes (Evans and Kim, 2013; Steptoe and Feldman, 2001) and the persistence of concentrated disadvantage in urban neighbourhoods (McAvay, 2020; Sampson, 2019). But there is little credibly causal evidence of the impact of chronic exposure to violence on educational outcomes. This paper connects these literatures by estimating whether – and how – the causal impact of violent crime on academic performance changes with repeated exposures. Repeated violence may sensitise children such that the impact on academic outcomes of an additional incident of violence is greatest for children in the most violent communities through increased emotional distress. Alternatively, repeated exposure may lead children to become desensitised to chronic violence, so the impact of an additional incident is weakest for children exposed to multiple violent crimes in their neighbourhoods.

Isolating this causal effect is difficult because disparities in academic performance between students from more and less violent communities may reflect residential sorting according to unobserved child or family characteristics or reflect unobserved stressors other than violence, such as persistent neighbourhood disadvantage or disinvestment that may create conditions for violence to emerge. We estimate the causal effect of chronic exposure to neighbourhood violent crime using two empirical strategies and detailed data on students and crime in NYC.

First, we use student-level longitudinal data to estimate value-added models linking student performance on standardised exams to violent crime. We compare test scores of children exposed to homicides or aggravated assaults on their residential block in the year prior to taking a standardised exam, with children living in the same census tract but on blocks with no violent incidents in that year. To shed light on the effects of repeated exposures, we observe how test scores change as the number of violent incidents on a child’s block increases. This strategy yields causal estimates if exposure to violence within census tracts is effectively random.

Second, we follow Sharkey (2010) and Sharkey et al. (2014) and use a regression discontinuity (RD) approach that exploits the timing of violent crimes relative to testing dates to estimate an acute effect on test scores. We compare children exposed in the week before the test to those exposed in the week after. We stratify the sample by the number of crime exposures in the year prior to the one-week window. If the timing of a crime relative to the test is random, this strategy yields a causal estimate of the acute effect of an additional crime exposure and shows how it varies with a student’s history of prior exposures.

To preview results, we find negative effects of exposure to violent crime on test scores that accumulate with exposure. Value-added models indicate exposure to violence in the year prior to testing lowers tests scores in English Language Arts (ELA) and Mathematics. Students exposed to violent crime three or more times in one-year score 0.016 standard deviations (SD) lower than students in their neighbourhood not exposed to crime, and 0.007 SD lower than those exposed to two crimes only. That is, performance decreases further with additional exposure supporting the sensitisation hypothesis among the full sample.

Our RD results show a significant acute effect for children living on blocks with multiple incidents of violence over the prior year, and no effect for children without prior exposures. Among students exposed to two or more crimes in the prior year, exposure to a recent violent crime decreases ELA scores by 0.04 SD. The negative effect is substantially larger for Black students exposed twice or more, amounting to a 0.08 SD decrease. These analyses provide strong evidence that children become sensitised to violent environments – in other words, the acute effect of violent crime increases with a student’s history of prior exposures.

Literature review

A rich body of work documents a negative link between violent environments and academic performance (Aizer, 2009; Bowen and Bowen, 1999; Burdick-Will et al., 2011; Caudillo and Torche, 2014; Delaney-Black et al., 2002; Hurt et al., 2001). Some of this work shows that additional violence exposures are linked with worse educational outcomes, suggesting students may become sensitised to violence. That said, with some exceptions, existing work is correlational, relying on small samples or using self-reported retrospective measures of crime exposure making it difficult to disentangle the effect of violent crime – and chronic violence – from other neighbourhood disadvantage.

Two key studies provide causal evidence of the negative consequences of crime in the short-run. Sharkey (2010) exploits the timing of homicides and assessment dates in Chicago, estimating the impact of exposure to homicides by comparing the performance of children living in the same neighbourhood but tested at different times. The results show that exposure to homicides less than a week before an assessment lower reading and vocabulary scores for Black students by 0.5 and 0.6 SD, respectively.

Sharkey et al. (2014) use a similar approach to isolate the acute effect of exposure to neighbourhood violent crime on standardised test performance for students in grades 3–8 attending NYC public schools and living in high poverty census tracts. They compare the test scores of students exposed to a violent crime one week before the test with the scores of children exposed one week after, under the assumption that the precise timing of neighbourhood violence is random. Results show that exposure to violence lowers ELA tests scores by 0.026 SD, largely due to the effect on Black students in elementary school.

Compared to Sharkey (2010), this paper uses a much larger sample (almost 40,000 observations) and focuses on state standardised exams. The analysis is restricted, however, to students living in high poverty census tracts which may limit generalisability. More importantly, the evidence on acute effects, raises – but leaves unanswered – questions about the effect of repeated violence as most children experience only one crime exposure within a one-week window. Are these acute effects larger for children that experience chronic violence?

Sensitisation versus desensitisation

To the extent that repeated crime exposures create additional stress for children, it is likely that academic outcomes suffer through increased emotional distress. Emotional distress may affect cognitive outcomes directly (McEwen and Sapolsky, 1995) or indirectly through behavioural changes such as difficulty sleeping (Heissel et al., 2018). Emotional distress from violence exposure can also lead to externalising or internalising behaviours that negatively correlate with education (Okano et al., 2020; Osofsky, 1999). In contrast, if chronic violence does not lead to worsening emotional distress after a certain number of exposures, then we may not see worsening academic performance among children within the most violent neighbourhoods. Whether and to what extent children’s emotional well-being is affected by chronic violence has been investigated in the psychological literature. This literature has developed theories of adaptation to violence and other environmental stressors.

The ‘sensitisation’ hypothesis (maladaptation) argues the cumulative toll of living in a violent environment may make children more sensitive to each additional exposure, suggesting that the impact of an additional incident on behavioural and cognitive outcomes should be greatest for children living in the most violent communities (Ng-Mak et al., 2004). Conversely, the ‘desensitisation’ hypothesis (adaptation), argues that children who are frequently exposed to crime may become desensitised to the effects of these events. The ‘desensitisation’ theory thus predicts that exposure to an additional violent incident will have a more modest, or even a negligible added effect on children living in more violent environments (Ng-Mak et al., 2004).

Existing studies have found support for both theories. Two studies focused on sixth graders suggest that additional violence exposures are not correlated with increased emotional distress suggesting desensitisation to violence. (Farrell and Bruce, 1997; Ng-Mak et al., 2004). That said, Ng-Mak et al. (2004), find that exposure to more crimes is associated with more aggressive behaviours. They refer to this mixed result as ‘pathologic adaptation’. In this case, more aggressive behaviours may negatively affect academic performance through increased disciplinary sanctions (Novak, 2021).

Other studies claim little support for the desensitisation hypothesis. For example, McCart et al. (2007) observe a positive linear relationship between more lifetime exposure to community violence and increased symptoms of post-traumatic stress among adolescents in a national sample, which they interpret as suggesting sensitisation.

In sum, existing work shows that exposure to neighbourhood violence affects behavioural and academic outcomes. However, the effects of repeated and past exposure – and whether children are sensitised or desensitised to violence – remain uncertain as most existing work yields mixed conclusions and is mostly correlational. In this paper, we provide causal estimates of the effects of exposure to violence on student outcomes and explore whether they differ by race and gender.²

Data

We use student level data from the NYC Department of Education and point specific crime data from the NYC Police Department from 2004 to 2010. These include the date, time and offense class of all crimes (except rape) during these years as well as the X and Y coordinates of each crime. We assign these crimes to specific blockfaces using shapefiles of streets. We connect each land parcel to a blockface using ArcGIS. A blockface consists of both sides of the street between two intersections (Supplemental Figure A1 in Appendix). This paper focuses on homicides and aggravated assaults,³ which are more serious and traumatic than other type of crimes.⁴ Exiting work has shown that local homicides are particularly detrimental for cognitive and behavioural outcomes (Caudillo and Torche, 2014; Sharkey, 2010; Sharkey et al., 2012) as well as homicides and aggravated assaults (Sharkey et al., 2014). These crimes are also more likely to be reported (Hart and Rennison, 2003) and attract attention (Brantingham and Uchida, 2021), thus residents of a blockface may be aware a homicide happened, even without direct exposure. We calculate that in each year of our sample, violent crimes account for 8%– 9% of reported crimes in NYC.

The education data contain individual-level records of all students enrolled for at least three years in NYC public schools in grades 3–8 between school years 2004/2005 and 2009/2010. The data include a rich set of demographic characteristics including race and ethnicity, gender, participation in special education, limited English proficiency, nativity, home language, eligibility for free or reduced-price lunch, grade level and test scores in ELA and math. Importantly, we know the building in which students reside in the fall semester, which we assign to a blockface to match to the crime data. We treat residential location as fixed through the end of the school year. We create crime exposure measures by counting the number of crimes on students’ blockfaces in a given window of time. While we cannot know whether a student has witnessed a crime, we label students ‘exposed’ if a violent crime has taken place on their residential blockface. We do not assume a student witnessed or was the victim of the crime. Because the blockface is a very small geographic unit, it is likely that a resident of a blockface would be aware of a serious offense such as a homicide or aggravated assault for example through increased police presence, sirens or discussion of the incident. That said, we are unable to determine who ultimately hears of an incident, and it is still possible some students are unaware of the crimes.

We use two analytic samples. Our value-added sample contains 1,264,113 student-year test score observations (382,489 unique students), distributed across 1181 schools and 2160 census tracts from grades 4–8.⁵ Over 40% of these students are exposed to violent crime in the year prior to testing (Table 1). Specifically, 18% are exposed to one violent crime, 9% to two violent crimes and 14% to three or more. Students exposed to violent crime score lower in both ELA and maths tests. For example, students exposed three or more times score, on average, 0.21 SD lower in ELA and 0.22 lower in maths than students with no exposures. Importantly, as exposure increases, student achievement declines. Further, Blacks and Hispanics are overrepresented among those repeatedly exposed to crime: 92% of students exposed to three or more violent crimes are Black or Hispanic. Students exposed to violence face greater educational challenges: larger percentages are limited English proficient and overage for grade, and over 90% are low income (defined by eligibility for free or reduced-price lunch).

Table 1.

Student characteristics by exposure to homicides and assaults, grades 4–8, years 2005–2010, ELA, prior year.

	Not exposed	Exposed
	Not exposed	Only once	Only twice	Three times more
Black	0.28 (0.45)	0.36 (0.48)	0.38 (0.49)	0.43 (0.49)
Hispanic	0.34 (0.47)	0.44 (0.50)	0.48 (0.50)	0.50 (0.50)
Asian	0.18 (0.38)	0.12 (0.33)	0.09 (0.28)	0.05 (0.22)
White	0.21 (0.41)	0.09 (0.28)	0.05 (0.22)	0.03 (0.16)
Female	0.50 (0.50)	0.51 (0.50)	0.51 (0.50)	0.52 (0.50)
Free/reduced price lunch	0.81 (0.39)	0.91 (0.29)	0.93 (0.25)	0.95 (0.22)
Foreign born	0.13 (0.33)	0.13 (0.34)	0.12 (0.33)	0.11 (0.31)
Home lang. not English	0.39 (0.49)	0.41 (0.49)	0.42 (0.49)	0.40 (0.49)
Special education	0.10 (0.31)	0.11 (0.31)	0.11 (0.31)	0.12 (0.32)
Limited English proficiency	0.04 (0.21)	0.06 (0.24)	0.08 (0.27)	0.08 (0.28)
Overage for grade	0.08 (0.27)	0.11 (0.32)	0.13 (0.33)	0.14 (0.35)
z_score ELA	0.16 (0.99)	−0.04 (0.92)	−0.12 (0.88)	−0.21 (0.84)
z_score MATHS	0.19 (0.98)	−0.03 (0.93)	−0.12 (0.91)	−0.22 (0.87)
Attendance rate	0.94 (0.07)	0.93 (0.07)	0.93 (0.08)	0.92 (0.08)
Observations = 1,264,113	739,350	232,016	115,057	177,377
Observations = 1,264,113	58.49%	18.35%	9.10%	14.03%

Note: Violent crimes include homicides and felony assaults that occurred on a student’s block in the year before the ELA test. Students are not exposed when the number of violent crimes on their block prior to the ELA test is equal to 0. Z_scores associated students exposed to one, two or three or more crimes with students not exposed to a crime within one year of the ELA and maths tests. Z_score for students not exposed to crime is their average across years and grades. Standard deviations in parentheses.

Our RD sample includes 37,041 student-year test score observations and 34,164 unique students. Table 2 presents descriptive statistics comparing students exposed before and after the ELA test, stratified by exposure prior to the one-week window. The majority of these students are Black or Hispanic. Half of the students with prior exposures are Hispanic and 40% are Black. Almost all students, regardless of their prior exposure, are low income.

Table 2.

Student characteristics by exposed to homicides and assaults, grades 4–8, 2005–2010, ELA.

	Exposed in one week window
	No previous exposure		Previous exposure
	Before	After	Before	After
Black	0.39 (0.49)	0.36 (0.48)	0.42 (0.49)	0.41 (0.49)
Hispanic	0.41 (0.49)	0.44 (0.50)	0.50 (0.50)	0.52 (0.50)
Asian	0.12 (0.32)	0.12 (0.32)	0.06 (0.23)	0.04 (0.20)
White	0.08 (0.28)	0.09 (0.28)	0.03 (0.17)	0.03 (0.17)
Female	0.51 (0.50)	0.52 (0.50)	0.51 (0.50)	0.51 (0.50)
Poor	0.90 (0.30)	0.90 (0.29)	0.95 (0.22)	0.95 (0.23)
Foreign born	0.15 (0.35)	0.13 (0.34)	0.13 (0.34)	0.12 (0.32)
Home lang. not English	0.41 (0.49)	0.41 (0.49)	0.42 (0.49)	0.42 (0.49)
Special education	0.11 (0.31)	0.10 (0.31)	0.11 (0.31)	0.11 (0.31)
Limited English proficiency	0.08 (0.27)	0.09 (0.28)	0.10 (0.30)	0.10 (0.30)
Overage for grade	0.10 (0.30)	0.11 (0.31)	0.14 (0.35)	0.15 (0.36)
Observations	4723	4622	12,589	15,107
Observations	50.5%	49.5%	45.5%	54.5%

Note: Column percentages. Students with no previous exposure are only exposed to a violent crime in the week before or after the ELA test. Students with previous exposures were exposed at least once in the year between ELA tests. Standard deviations in parentheses. ELA: English Language Arts

Empirical strategy

Value added models

We begin by estimating value added regression models linking student performance on standardised tests to violent crime exposure:

T e s t_{i c t} = \begin{matrix} α_{t} + β_{1} O n e_{i c t} + β_{2} T w o_{i c t} \\ + β_{3} T h r e e_{i c t} + λ T e s t_{i c t - 1} + X'_{i c t} θ \\ + γ_{g} + δ_{c} + ε_{i c t} \end{matrix}

(1)

In this specification, test represents student i’s test score on a standardised test (ELA or maths), measured as z scores standardised for each grade citywide, with a mean of zero and a standard deviation of one; c indexes census tracts, and t indexes time. In this model, One equals one if i was exposed to one violent crime only in the year between test dates or inter-test year⁶– and it is zero otherwise. Accordingly, Two equals one if i was exposed to two crimes only, and Three equals one if i was exposed three times or more in the year before a standardised test. The impact of violent crime is identified by comparing the performance of two otherwise similar students – one living on a block on which a violent crime occurred; the other living in the same census tract, but on a block with no violent crime in the year prior to testing.

The coefficients of interest – $β_{1}, β_{2}$ and $β_{3}$ – capture the effect of violent crime on test scores for different number of exposures and will be unbiased if the precise location of violent crimes across blockfaces within census tracts is uncorrelated with unobserved variables that would affect student performance. Comparing these coefficients sheds light on the marginal effect of additional exposure to crime, and the sensitisation (versus desensitisation) hypotheses. If students become sensitised to neighbourhood crime this marginal effect should increase with additional exposures (i.e. the difference between $β_{1} and β_{2}$ would be smaller than the difference between $β_{2}$ and $β_{3}$ ). Alternatively, if students become desensitised, the marginal effect should decrease with additional exposures. The remaining variables are as follows: $λ T e s t_{i c t - 1}$ are i’s test scores lagged one year, $X'$ is a vector of student demographic controls, $α_{t}, γ_{g}$ and $δ_{c}$ are year, grade and census tract fixed effects, respectively. An alternative specification replaces census tract fixed effects with school fixed effects to account for time-invariant characteristics of the school environment. That said, census tract fixed effects are our preferred specifications because they capture specific elements of localised neighbourhoods that might affect both exposure to violence and test scores. High population density in NYC means the roughly 2000 census tracts in the city are geographically quite small. In fact, the average NYC elementary school zone contains just six census tracts.⁷

Regression discontinuity design

Our second approach is an RD model that exploits variation in the timing of homicides and aggravated assaults relative to testing dates. This strategy compares students exposed to violent crime in the week before the test with students exposed in the week after.⁸ We estimate:

T e s t_{i t} = α_{t} + β C r i m e_{i t} + X'_{i t} θ + γ_{g} + ε_{i t}

(2)

In this model test is still the outcome of interest, and Crime equals 1 if student i was exposed to a violent crime in the week before the test and it is 0 if exposure happened in the week after. $β$ is the coefficient of interest. The model includes year ( $α_{t}$ ) and grade ( $γ_{g}$ ) fixed effects, and a vector of student controls (X′). We restrict the sample to students exposed to violent crime on their blockface either the week before or after the test, and exclude students exposed both before and after in a given year.

This approach should yield an unbiased estimate of the acute effect if the timing of the crime relative to the test is effectively random. As shown in Table 2 students exposed before and after are similar by race, gender and poverty status. The samples also look fairly similar on a broader set of demographic characteristics including participation in special education, limited English proficiency and nativity. To further establish the similarity between the treatment and comparison groups we estimate a series of regressions of each demographic characteristic on the crime exposure dummy that equals 1 if a student was exposed to a homicide or assault in the week before the test, and 0 if exposure happened the week after. We conduct this test for all students, and we also stratify the sample by exposure in the prior year. Results from these regressions – reported on Supplemental Table A1 – provide further evidence that the samples are balanced on demographic controls.⁹

Incorporating prior exposure

Our key aim is to identify how the impact of an additional crime exposure changes with more exposure. To do so, we estimate equation (2) stratifying the sample by the number of violent crimes on a student’s block in the year prior to the one-week window (Supplemental Figure A2). We define one year as the period between test dates, and to calculate previous exposure we count the number of crimes on a student’s block in the year prior to the test minus the crimes that occurred in the week right before the test. In this way, if a student was not exposed prior to the one-week window, $β$ captures the acute effect of the first exposure. If a student was exposed once before the one-week window, $β$ identifies the impact of the second exposure. For students exposed to two violent crimes in the prior year, this coefficient identifies the acute effect of the third crime exposure.

The interpretation of these coefficients depends on whether there is selection into the prior exposure categories. To test whether such selection exists, we estimate a series of separate binary regressions of each demographic characteristic on four crime exposure dummy variables (no prior exposure, one exposure, two exposures or three or more previous exposures; 44 regressions in total). Results from these regressions show little evidence of selection into one of these crime categories within census tracts (Supplemental Table A2). In all specifications discussed in this paper, standard errors are clustered at the census tract level.

Results

Table 3 reports ELA results for value-added specifications. The negative effect of violent crime on test scores increases with the number of exposures, with a larger effect for students exposed three or more times within a year compared to those not exposed. Specifically, for the approximately 14% of our sample exposed three or more times, test score losses range from 0.04 SDs in models that include demographic controls only (column 2) to 0.016 in models with full controls (column 4). In our preferred specification with census tract fixed effects (column 4), test scores for students exposed to one crime are 0.007 SD lower on ELA (17.5% decline over the mean for this group), scores for students exposed to two crimes are 0.009 SD lower, and scores for those with three or more exposures decline by 0.016 SD (roughly a 7.5% decline over the mean for these groups). The difference between these coefficients shows that the marginal effect of an additional exposure increases from −0.002 to −0.007 SD. These results provide some support for the sensitisation hypothesis. We also estimated linear and quadratic specifications with similar results. The marginal effect of exposure is a decline in test scores of 0.003−0.005 SD. Quadratic models show little evidence of desensitisation (Supplemental Table A3)

Table 3.

Value-added results, ELA, one year.

DV: ELA z-score	(1)	(2)	(3)	(4)
One crime	−0.204** (0.009)	−0.020** (0.002)	−0.009** (0.002)	−0.007** (0.002)
Two crimes	−0.283** (0.010)	−0.030** (0.003)	−0.012** (0.002)	−0.009** (0.002)
Three or more crimes	−0.373** (0.011)	−0.042** (0.003)	−0.021** (0.002)	−0.016** (0.002)
Student controls	No	Yes	Yes	Yes
School FX (1137)	No	No	Yes	No
Census tract FX (2153)	No	No	No	Yes
Observations	1,264,113	1,264,113	1,264,113	1,264,113
R ²	0.024	0.504	0.517	0.509

Note: Student controls include: female, Black, Hispanic, Asian, poor, special education, foreign born, home language not English, limited English proficiency, overage for grade and test scores lagged one year. All models include year and grade fixed effects. Standard errors are clustered at the census tract level. Crime includes homicides and aggravated assaults that happened in the year between ELA tests. Sample includes students in grades 4–8 between AY 2004/2005 and 2009/2010. Standard errors in parentheses.

FX: Fixed effects; DV: Dependent variable.

p < 0.01.

We then investigate differences by race and ethnicity, and gender. We find that Black students exposed to violent crime score consistently lower on ELA with the largest test score losses for those exposed three or more times compared to those not exposed (0.017 SD – marginal effect is −0.006 SD; Table 4, column 1). This coefficient represents 13% of the Black-White test score gap in this sample. Hispanic students seem to only be affected by higher levels of violence, scoring 0.011 SD lower in ELA when exposed to three or more violent crimes (11% of the Hispanic-White achievement gap in this sample).¹⁰ Table 4 suggests that White students are particularly sensitive to violence, scoring 0.03 SD lower when exposed to two crimes. We should note, however, that there are very few White students exposed to such levels of violence. White students comprise only 6151 of the 115,098 students exposed twice (5.3%), and only 4500 of the 177,464 students exposed to violent crime three times or more (2.5%). As for gender, girls exposed to three or more violent crimes suffer a larger reduction in test scores than boys exposed to similar levels of violence (coefficients are 0.022 and 0.011 SD, respectively), though there are no differences at lower levels of exposure.

Table 4.

Value-added results, ELA, one year, race and ethnicity and gender.

DV: ELA z-score	Black	Hispanic	Asian	White	Boys	Girls
DV: ELA z-score	(1)	(2)	(3)	(4)	(5)	(6)
One crime	−0.005⁺ (0.003)	−0.004⁺ (0.003)	−0.010⁺ (0.006)	−0.010 (0.007)	−0.007** (0.002)	−0.007** (0.003)
Two crimes	−0.011** (0.003)	−0.002 (0.003)	−0.010 (0.009)	−0.030** (0.010)	−0.007* (0.003)	−0.011** (0.003)
Three or more crimes	−0.017** (0.003)	−0.011** (0.003)	−0.015 (0.010)	−0.015 (0.014)	−0.011** (0.003)	−0.022** (0.003)
Student controls	Yes	Yes	Yes	Yes	Yes	Yes
Census tract FX	Yes	Yes	Yes	Yes	Yes	Yes
Observations	406,982	495,534	177,732	183,865	622,947	641,166
R ²	0.470	0.485	0.428	0.423	0.511	0.506

Note: Student controls include: female, Black, Hispanic, Asian, poor, special education, foreign born, home language not English, limited English proficiency and overage for grade. All models include year, grade fixed effects and test scores lagged one year. Standard errors are clustered at the census tract level. Crime includes homicides and aggravated assaults in the year between ELA tests. Sample includes students in grades 4–8 between AY 2004/2005 and 2009/2010. Standard errors in parentheses.

p < 0.01. *p < 0.05. ⁺p < 0.1.

While these analyses suggest that exposure to crime modestly lowers ELA scores and children become sensitised to violence, it is not clear that these estimates warrant a causal interpretation – in particular, they may reflect unobserved differences between students exposed to violence and those not. Thus, we turn to a regression discontinuity design.

The acute effect of an additional crime exposure

Table 5 presents RD results for our baseline specification and stratified by the number of previous exposures. On average, exposure to a homicide or felony assault in the week before the test lowers ELA test scores by 0.025 SD compared to students exposed the week after (column 1). Some students are exposed to more than one violent crime prior to the one-week window. Indeed, 40.5% of students exposed in a one-week window were also exposed to three or more crimes in the previous year, and only 25% were not exposed prior to the one-week window. Is the acute effect larger for children with more prior exposures? Results in columns 2–7 suggest ‘yes’. Students exposed to two or more crimes in the prior year score 0.038 SD lower after the additional exposure in the week before the ELA test relative to students with similar prior exposure but exposed in the week after the test (column 7). Conversely, students without previous exposures or with one prior exposure score no lower after being exposed to a violent crime. In sum, the acute effect of violent crime in column 1 is driven by students with two or more prior crime exposures. Note that effects of additional crime exposures are only statistically significant for students with three or more prior exposures (column 5). These results suggest that children become sensitised to violence. The impact of an additional exposure to violent crime is greatest for children exposed to higher levels of violence over the course of the prior year.¹¹

Table 5.

RD results, ELA, one week window.

DV: ELA z-score	Previous exposure
	All	None	One	Two	Three or more	None or one	Two or more
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Crime	−0.025* (0.011)	0.013 (0.021)	−0.008 (0.023)	−0.046 (0.028)	−0.033* (0.015)	−0.003 (0.016)	−0.038** (0.013)
Student controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	37,041	9345	7530	5147	15,019	16,875	20,166
R ²	0.198	0.221	0.190	0.192	0.173	0.209	0.179

Note: Student controls include: female, Black, Hispanic, Asian, poor, special education, foreign born, limited English proficiency, home language not English, overage for grade. All models include year, and grade fixed effects. Standard errors are clustered at the census tract level. Students exposed both before and after the test are excluded. Crime includes homicides and aggravated assaults. Sample includes students in grades 4–8 between AY 2004/2005 and 2009/2010. Standard errors in parentheses. **p < 0.01. *p < 0.05.

RD: Regression Discontinuity

Table 6 shows results for Black and Hispanic students. We restrict our analysis to these two groups because each is overrepresented in our sample, while there are very few White and Asian students. We find the largest negative acute effect for Black students exposed twice or more in the prior year (0.08 SD lower in ELA). Consistent with prior research (Sharkey et al., 2014), we find no acute effect for Hispanic students.¹² Since the number of Black and Hispanic students exposed two or more times in the prior year is similar, this disparity cannot be simply attributed to differences in violence exposure, and it may suggest other differences in the neighbourhoods or schools of Black and Hispanic students, or in how they cope with violence exposure. Table 6 shows no differences by gender. The negative acute effect is 0.04 SDs for girls and boys exposed twice or more in the prior year.

Table 6.

RD results, ELA one week window, race/ethnicity and gender.

DV: z-score ELA	Previous exposure
	None	One	None or one	Two or more	None	One	Noneor one	Two or more
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Race/Ethnicity
Crime × Black	−0.028(0.033)	−0.024 (0.037)	−0.035(0.024)	−0.078** (0.019)
Crime × Hispanic	0.049⁺(0.027)	−0.017(0.026)	0.010(0.019)	−0.011(0.017)
Gender
Crime × Boys					0.016 (0.026)	−0.006(0.031)	0.000 (0.020)	−0.039* (0.018)
Crime × Girls					0.011(0.029)	−0.010(0.030)	−0.004 (0.021)	−0.038* (0.017)
Student controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
F-Stat	3.64	0.02	2.59	7.22	0.02	0.01	0.03	0.00
Prob > F	0.0566	0.8766	0.1080	0.0074	0.8793	0.9197	0.8619	0.9736
Observations	7468	6609	14,077	18,908	9345	7530	16,875	20,166
R ²	0.153	0.153	0.152	0.157	0.222	0.190	0.209	0.179

Note: Student controls include: female, Black, Hispanic, poor, special education, foreign born, limited English proficiency, home language not English and overage for grade. All models include year, and grade fixed effects. Standard errors are clustered at the census tract level. Students exposed both before and after are excluded. Crime includes homicides and aggravated assaults. Sample includes students in grades 4–8 between AY (Academic Year) 2004/2005 and 2009/2010. Standard errors in parentheses.

p < 0.01. *p < 0.05. ⁺p < 0.1.

Robustness tests

We test the robustness of our results in multiple ways. We first estimate RD models using a two-week exposure window, expanding the analytic sample substantially. Results are substantively unchanged. The negative acute effect is driven by children with prior exposures, and those with two or more previous exposures specifically (Supplemental Table A5).¹³ Subgroup results are also robust to the larger sample. The acute effect is negative and large for Black students and for girls exposed twice or more in the prior year (Supplemental Table A6).

Second, we estimate models with controls for property crimes and robberies (another type of violent crime), as well as the interaction of these crime variables with homicides and aggravated assaults (Supplemental Table A7). In these specifications, the negative effect of exposure to homicides and aggravated assaults on ELA test scores persists in magnitude and significance (column 1). Effects are still driven by students with two or more prior exposures (column 3). Exposure to robberies or property crimes does not have an independent effect, nor does it moderate the effect of homicides and aggravated assaults for students previously exposed to violent crime.¹⁴ We also estimated value-added models adding additional indicators for exposure to one, two or three or more robberies. Our results for exposure to homicides and aggravated assaults remain unchanged with these additional variables. Point estimates for exposure to robberies are statistically significant but slightly smaller (Supplemental Table A8).

Third, we estimated models for maths test scores. The value-added yields similar results. The negative effect of crime exposure is 0.01, and it increases with the number of exposures (Supplemental Table A9). Results for race and ethnicity, and gender subgroups are also similar. Black students exposed to crime score consistently lower on maths exams regardless of crime exposure, while Hispanic, Asian and White students seem to be affected by higher levels of exposure only (Supplemental Table A10). In sum, value-added results show negative effects of violent crime on both ELA and maths. In contrast, the RD models show no acute effect of exposure to neighbourhood crime on maths test scores – due, perhaps, to better controlling for unobserved differences in students than the value-added models (Supplemental Table A11).

Conclusion

This paper investigates the effect of repeated exposure to neighbourhood violent crime on student performance using two empirical approaches. Consistent with prior work, we find that students exposed to violent crime perform worse on reading tests. More importantly, we find that chronic neighbourhood violence is especially harmful. Children repeatedly exposed to violent crime over a one-year period score lower on standardised tests, and the marginal effects of added crime grow with the number of exposures. Our estimates also suggest that chronic violence exacerbates the acute effect of crime. That is, students who experience violent crime on a regular basis become more sensitised to violence than students for whom neighbourhood crime may be an isolated event and their academic performance suffers accordingly. That is, the negative effect of community violence builds with repeated exposure, potentially resulting in lasting deficits in academic performance for children living within the most violent urban neighbourhoods. The magnitude of these estimates is comparable to the positive findings in test scores associated with crime reductions (Torrats-Espinosa, 2020). Further, our effects are comparable to the gains in test scores of class size reductions (Cho et al., 2012) suggesting neighbourhood violence has the potential to undermine the positive gains from some educational interventions. Schools and teachers may pay particular attention to helping students repeatedly exposed to neighbourhood violence for example by fostering a welcoming and safe school climate (Laurito et al., 2019).

Our results also shed light on the role of repeated neighbourhood violence in exacerbating educational disparities. Black students disproportionately experience repeated exposures, and the effect widens the Black-White test score gap. The acute effect for Black students with two or more prior exposures amounts to 17% of the estimated Black-White test score gap for this group.¹⁵

It is possible that children exposed to more violent crimes on their block are also more likely to have witnessed a crime or know someone who is the victim of a crime. If witnessing a crime has a worse effect on academic performance than hearing about a crime, this difference could drive some our results for chronic exposure. Unfortunately, we are unable to know whether children witness crimes, and so we cannot identify these specific mechanisms. That said, our results highlight that living in violent neighbourhoods has effects beyond direct exposure that can negatively impact academic performance. This finding supports the idea the effect of neighbourhood violence on individual outcomes extends beyond victimisation and witnessing (Sharkey, 2018) and aligns with studies that show community-wide violent events affect academic outcomes even if not directly experienced (Gershenson and Tekin, 2018).

We find the largest effects on ELA, and smaller or no effects on maths. This finding is consistent with previous research that shows neighbourhood violence negatively affects the development of language skills and performance on reading tests (Sharkey et al., 2014; Torrats-Espinosa, 2020). One possible explanation is that performance in maths and reading responds to different cognitive and self-regulatory mechanisms. For example, Sharkey et al. (2012) found that exposure to homicides lowered attention, and impulse control. Evidence from psychology suggests that impulse control may be particularly important for reading instruction and the development of reading skills, while maths instruction, which usually involves more individual work, may require other skills such as self-monitoring (Liew et al., 2008).

Consistent with findings in Sharkey et al. (2014) we find no impact of violent crime for Hispanic students, which remains something of a puzzle considering that Black and Hispanic students in our sample are exposed to similar levels of violence. These varying results may reflect other differences in the neighbourhood and school contexts of Black and Hispanic students that may moderate the impact of community violence and affect their coping strategies that future work should investigate.

Finally, there are some limitations to the present study. First, while we isolate the effect of neighbourhood violence on test scores, our study does not speak to the many processes of neighbourhood disadvantage that may contribute to violence. As such, our paper is not able to identify neighbourhood-level mechanisms that may explain our findings. Future research should explore these interconnections. Second, we focus on test performance as the main outcome of interest. Performance on standardised tests, albeit important in a world of high stakes testing, is not the only outcome that can be affected by exposure to violent crime. Exposure to violent crime can also affect school attendance, by making children more fearful or shaping their socio-emotional well-being. To the extent that violent crime affects externalising behaviours, we may see worse disciplinary outcomes among students exposed to violence. Future work should investigate the effect of violent crime on students’ disciplinary sanctions and examine whether racial and ethnic differences in exposure to violence help to explain some of the observed racial and ethnic disparities in disciplinary outcomes in US schools. Investigating the effect of neighbourhood crime on these other outcomes would provide a more comprehensive view of how neighbourhood violence affects all aspects of child well-being and may contribute to exacerbate existing disparities in education.

Supplemental Material

sj-docx-1-usj-10.1177_00420980211052149 – Supplemental material for The academic effects of chronic exposure to neighbourhood violence

Supplemental material, sj-docx-1-usj-10.1177_00420980211052149 for The academic effects of chronic exposure to neighbourhood violence by Amy Ellen Schwartz, Agustina Laurito, Johanna Lacoe, Patrick Sharkey and Ingrid Gould Ellen in Urban Studies

Footnotes

Acknowledgements

We gratefully acknowledge the generous support of the William T Grant Foundation. We also thank Meryle Weinstein, the Furman Center for Real Estate and Urban Policy at New York University for facilitating the crime data and three anonymous reviewers who provided valuable comments and feedback.

Funding

This research was funded by grant No180144 ‘Crime, Context and Children’s Academic Performance’, William T. Grant Foundation.

ORCID iD

Agustina Laurito

Supplemental material

Supplemental material for this article is available online.

Notes

References

Aizer

(2009) Neighborhood violence and urban youth. In: Gruber

(ed.) The Problems of Disadvantaged Youth: An Economic Perspective. Chicago, IL: University of Chicago Press, pp.275–307.

Bowen

(1999) Effects of crime and violence in neighborhoods and schools on the school behavior and performance of adolescents. Journal of Adolescent Research 14(3): 319–342.

Brantingham

Uchida

(2021) Public cooperation and the police: Do calls-for-service increase after homicides? Criminal Justice Journal 73: 101785.

Burdick-Will

Ludwig

Raudenbush

, et al. (2011) Converging evidence for neighborhood effects on children’s test scores: An experimental, quasi-experimental, and observational comparison. In: Duncan

Murnane (eds) Whither Opportunity? Rising Inequality, Schools, and Children’s Life Chances. New York, NY: Russell Sage Foundation, pp.255–276.

Caudillo

Torche

(2014) Exposure to local homicides and early educational achievement in Mexico. Sociology of Education 87(2): 89–105.

Cho

Glewwe

Whitler

(2012) Do reductions in class size raise students’ test scores? Evidence from population variation in Minnesota’s elementary schools. Economics of Education Review 31(3): 77–95.

Delaney-Black

Covington

Ondersma

, et al. (2002) Violence exposure, trauma, and IQ and/or reading deficits among urban children. Archives of Pediatrics & Adolescent Medicine 156: 280.

Evans

Kim

(2013) Childhood poverty, chronic stress, self-regulation, and coping. Child Development Perspectives 7(1): 43–48.

Farrell

Bruce

(1997) Impact of exposure to community violence on violent behavior and emotional distress among urban adolescents. Journal of Clinical Child Psychology 26(1): 2–14.

10.

Gershenson

Tekin

(2018) The effect of community traumatic events on student achievement: Evidence from the beltway sniper attacks. Education Finance and Policy 13(4): 513–544.

11.

Hart

Rennison

(2003) Reporting Crime to the Police, 1992-2000. Washington, DC: U.S. Department of Justice, Office of Justice Programs.

12.

Heissel

Sharkey

Torrats-Espinosa

, et al. (2018) Violence and vigilance: The acute effects of community violent crime on sleep and cortisol. Child Development 89(4): e323–e331.

13.

Hurt

Malmud

Brodsky

, et al. (2001) Exposure to violence. Psychological and academic correlates in child witnesses. Archives of Pediatrics & Adolescent Medicine 155(12): 1351–1356.

14.

Laurito

Lacoe

Schwartz

, et al. (2019) School climate and the impact of neighborhood crime on test scores. RSF: The Russell Sage Foundation Journal of the Social Sciences 5(2): 141–166.

15.

Liew

McTigue

Barrois

, et al. (2008) Adaptive and effortful control and academic self-efficacy beliefs on achievement: A longitudinal study of 1st through 3rd graders. Early Childhood Research Quarterly 23: 515–526.

16.

McAvay

(2020) Socioeconomic status and long-term exposure to disadvantaged neighbourhoods in France. Urban Studies 57(13): 2663–2680.

17.

McCart

Smith

Saunders

, et al. (2007) Do urban adolescents become desensitized to community violence? Data from a national survey. American Journal of Orthopsychiatry 77(3): 434–442.

18.

McEwen

Sapolsky

(1995) Stress and cognitive function. Current Opinion in Neurobiology 5(2): 205–216.

19.

Ng-Mak

Salzinger

Feldman

, et al. (2004) Pathologic adaptation to community violence among inner-city youth. American Journal of Orthopsychiatry 74(2): 196–208.

20.

Novak

(2021) Trajectories of exclusionary discipline: Risk factors and associated outcomes. Journal of School Violence 20(2): 182–194.

21.

Okano

Jeon

Crandall

, et al. (2020) The cascading effects of externalizing behaviors and academic achievement across developmental transitions: Implications for prevention and intervention. Prevention Science 21(2): 211–221.

22.

Osofsky

(1999) The impact of violence on children. The Future of Children 9(3): 33–49.

23.

Rasmussen

Aber

Bhana

(2004) Adolescent coping and neighborhood violence: Perceptions, exposure, and urban youth’s efforts to deal with danger. American Journal of Community Psychology 33(1–2): 61–75.

24.

Sampson

(2019) Neighbourhood effects and beyond: Explaining the paradoxes of inequality in the changing American metropolis. Urban Studies 56(1): 3–32.

25.

Sharkey

(2010) The acute effect of local homicides on Children’s cognitive performance. Proceedings of the National Academy of Sciences of the United States of America 107: 11733–11738.

26.

Sharkey

(2018) The long reach of violence: A broader perspective on data, theory, and evidence on the prevalence and consequences of exposure to violence. Annual Review of Criminology 1: 85–102.

27.

Sharkey

Schwartz

Ellen

, et al. (2014) High stakes in the classroom, high stakes on the street: The effects of community violence on students’ standardized test performance. Sociological Science 1: 199–220.

28.

Sharkey

Tirado-Strayer

Papachristos

, et al. (2012) The effect of local violence on children’s attention and impulse control. American Journal of Public Health 102(12): 2287–2293.

29.

Steptoe

Feldman

(2001) Neighborhood problems as sources of chronic stress: Development of a measure of neighborhood problems, and associations with socioeconomic status and health. Annals of Behavioral Medicine 23(3): 177–185.

30.

Torrats-Espinosa

(2020) Crime and inequality in academic achievement across school districts in the United States. Demography 57(1): 123–145.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.17 MB