The Effect of 311 Calls for Service on Crime in D.C. at Microplaces

Abstract

This study tests the broken windows theory of crime by examining the relationship between 311 calls for service and crime at the street segment and intersection level in Washington, D.C. Controlling for a set of micro-level covariates as well as unobserved neighborhood-level effects using negative binomial regression models, it is found that detritus- and infrastructure-related calls for service have a positive, but small effect on crime. The results suggest that 311 calls for service are a valid indicator of physical disorder where available, and the findings partially confirm the broken windows theory. Given the small effects though, reducing physical disorder is unlikely to result in appreciable declines in crime.

Keywords

311 broken windows microplace street units physical disorder

Introduction

This article examines the relationship between 311 calls for service and crime in Washington, D.C., at the geographic level of street segments and intersections. The 311 number was reserved nationwide in the United States in 1997 as a service where citizens could voice complaints about physical disorder, and was originally intended to be a resource for community policing strategies to identify areas of needed service (Community Oriented Policing Services [COPS], 2007). The broken windows theory of crime would suggest that reducing disorder should, in the long run, reduce crime (Wilson & Kelling, 1982).

There are two motivations for this current work. One is that scholars interested in measuring disorder have previously used data collections via surveys or systematic social observation. The 311 number is a potential source of physical disorder in the environment that can potentially replace costly systematic social observation collections, or used in lieu of surveys of citizen perceptions. This allows one to measure disorder at relatively small places and regularly over time.

The second motivation is that prior work that has examined 311 calls for service (or similar databases) and crime has examined the relationship at the census tract or block group level (Boggess & Maskaly, 2014; O’Brien & Sampson, 2015; Yang, 2010, but see O’Brien & Winship, 2016, for a recent exception). The theory behind the relationship between physical disorder and crime in broken windows theory is contingent upon seeing physical disorder. As such, examining the relationship at smaller microunits has the advantage to be able to test that the relationship between disorder and crime is actually limited to the immediate visual arena.

Prior work using surveys of individuals questions whether people are unbiased in their assessments of disorder, and that perceptions may be influenced by stereotypes of particular neighborhoods (Sampson & Raudenbush, 2004). To test for the priming effects of physical disorder independent of such neighborhood effects, this article considers microplace street units as nested within neighborhoods and uses fixed effects regression models. Also included are covariates at the micro-level that signify whether a location is a busy place, such as a shopping mall or a bar, which may be confounded with physical disorder (Wilcox, Quisenberry, Cabrera, & Jones, 2004).

Broken Windows Theory and Measuring Disorder

Broken windows theory was introduced by Wilson and Kelling (1982) to explain why particular urban neighborhoods had elevated crime problems. If someone sees a broken window, a visible cue of disorder, this signals to an individual that no one cares, and one can subsequently break more windows without repercussion. Wilson and Kelling took an experiment from Zimbardo (1973) as the motivation for this observation, in which Zimbardo placed an abandoned car in two locations, one nearby the New York University campus in the Bronx, New York City, and the other nearby the Stanford campus in Palo Alto, California. The car in the Bronx was stripped of parts and vandalized almost immediately, whereas the car in Palo Alto was left alone (until Zimbardo himself started to damage the vehicle).

To test this hypothesis, some scholars have measured disorder via surveys.¹ For example, survey items from Swatt, Varano, Uchida, and Solomon (2013) are questions asking respondents to rate “Dirty or unkempt buildings and lots” and “Graffiti” (for physical disorder) and “Homeless loitering” and “Groups of young people hanging out/around” (for social disorder) on an ordinal scale as being no problem, some problem, or big problem (p. 10).

The main critique of survey measures of disorder are that individual perceptions of disorder are potentially biased by resident stereotypes of particular neighborhoods or people (Sampson & Raudenbush, 2004; Yang & Pao, 2015), biased by the characteristics of respondents (Hipp, 2010; Wallace, 2015; Wallace, Louton, & Fornango, 2015), or that citizens simply cannot disentangle the difference between social disorder and crime itself (Gau & Pratt, 2008).² If it is the case that citizens cannot tell the difference between social disorder and crime itself, it suggests that using citizen-based measures of disorder in cross-sectional research to test broken windows theory is problematic (Hinkle, 2013; Link, Kelly, Pitts, Waltman-Spreha, & Taylor, 2014).

This creates a particular problem for broken windows theory, if individuals cannot perceive the difference between disorder and crime, how can disorder affect behavior independent of crime itself? First, one need not be consciously aware of particular stimuli for it to effect an individual’s behavior, see much of the psychological literature on priming (Kahneman, 2011). Second, the majority of the survey literature testing broken windows theory attempts to test the link between disorder and fear of crime. Wilson and Kelling’s original theory would be largely unchanged if they also associated physical and/or social disorder cues directly with more serious criminal behavior. For one example, Wilson and Kelling (1989) give in a later article,

A rash of burglaries may occur because drug users have found a back alley or an abandoned building in which to hang out. In their spare time, and in order to get money to buy drugs, they steal from their neighbors. If the back alleys are cleaned up and the abandoned buildings torn down, the drug users will go away (p.47).

As such, there is still utility in seeing whether physical disorder leads to more serious lawbreaking behavior, even if the link between disorder and fear of crime is in doubt.

Systematic social observation of places is another source of information that researchers have used to measure disorder (Raudenbush & Sampson, 1999; Sampson & Raudenbush, 1999). For example, Sampson and Raudenbush (1999) used measures of the presence or absence of cigarettes, litter, empty beer bottles, graffiti, abandoned cars, condoms, and syringes. These measures have the advantage that they are objective measures of disorder, as opposed to subjective measures based on resident perceptions. Studies using measures based on systematic social observation of places tend to find little evidence of the relationship between physical disorder and crime when controlling for other sociodemographic characteristics of neighborhoods (Harcourt & Ludwig, 2006; Sampson & Raudenbush, 2004).

In general, 311 calls are an administratively collected sign of physical disorder. Two prior examples using administratively collected data sets of disorder are a series of work using reports of physical disorder to public works in Seattle (Weisburd, Groff, & Yang, 2012, 2014; Yang, 2010) and the work of O’Brien and colleagues examining 311 calls in Boston and crime (O’Brien & Sampson, 2015; O’Brien, Sampson, & Winship, 2015; O’Brien & Winship, 2016). Where a specific 311 line is not in place, some physical disorder complaints are directed at the police department and are available in computer automated dispatch (CAD) data (Boggess & Maskaly, 2014; Wallace, Hedberg, & Katz, 2012). For example, Boggess and Maskaly (2014) used CAD calls for “intoxicated, unwanted, or undesirable persons, graffiti, abandoned vehicles, litter, illegal dumping, and suspicious persons, vehicles, and circumstances” (p. 173) as measures of both physical and social disorder (as opposed to misdemeanor arrests). In fact, this was the original motivation for a 311 line to begin with.

A History of 311 Calls for Service

Many calls for service the police receive are not related to law enforcement, but oriented to issues of public disorder (Bittner, 1967). In 1997, in the United States, per a Community Oriented Policing Office (COPS, a part of the United States Department of Justice) request, the Federal Communications Commission reserved the 311 number nationally for nonemergency calls (COPS, 2007). The program was intended to both reduce calls to 911 (Rogers, 1999) as well as provide a source of information related to quality of life issues that a community-oriented policing strategy might tackle.

Thus, 311 calls for service are a potential measurement source of physical disorder, and by construction were to be of use in police for identifying physical disorder problems to address in the community. Being a regularly collected administrative data set has the benefit over physical street audits, in that collection of the data is routine and covers the entire city. But, because 311 data are self-reported, the same as official crime, we may anticipate some of the same biases (Klinger & Bridges, 1997). For one example, the city of Boston created a cell phone application to report potholes, with the result that minority and elderly populations were severely underrepresented (Executive Office of the President, 2014). In an evaluation of 311 calls in Boston, O’Brien, (2015) found that residents who owned a home were more likely to phone in a complaint. But in another evaluation, O’Brien et al. (2015) determined that 311 calls provide a reliable measure of disorder at the block group level, using a physical audit to validate the measure. At the more micro-level of addresses, O’Brien and Winship (2017) found a strong correlation between 311 calls for service and crime in subsequent years. So, although one should be wary of potential biases when using self-report data (both for 311 calls for service and crime), there is prior evidence for the validity of using 311 calls for service data as a measure of disorder.

Identifying the Effect of 311 Calls for Service on Crime

Because the theory behind physical disorder is based on visibly seeing disorder, a stringent test will show that disorder only affects crime in the immediate visual area. To conduct this test, I differentiate the local effect of counts of physical disorder on crime, as well as the spatial effects of the first- and second-order spatial lag of counts of physical disorder on crime. For a simplified example, I examine the correlation between 311 calls for service on 1st and Main with counts of crime on 1st and Main. Then, I examine the correlation between calls for service on neighboring streets, such as 2nd and Main, with counts of crime on 1st and Main. Disorder on 2nd and Main should not be visible to those on 1st and Main, and so should have no effect on crime per broken windows theory.

Due to having to make discrete spatial units, it is possible for disorder to be visible within the first-order neighbors just across the border between any two street units. But by the time one gets to second-order neighbors, one is frequently over a block away, and so conditional on other control variables such effects should be zero. If disorder affects crimes further away than can be seen, it suggests there are omitted variables correlated both with physical disorder and crime not included in the model.

In addition to these spatial lags of disorder, this study differentiates between detritus-related aspects of physical disorder, such as illegal dumping and yard waste, as these reflect transient measures of disorder, and infrastructure-related signs of disorder, such as a potholes and sidewalk repair, as these are more permanent (Taylor, 2001). There are two reasons for this distinction, one theoretical and one related to the specific research design.

Broken windows theory makes no distinction between different types of observable physical disorder. A broken window or garbage on the street both provide a signal to potential offenders that informal social control is likely minimal in a particular area. But, different types of physical disorder vary in their persistence. In the case of detritus-related disorder, one motivated individual could theoretically take care of the problem in a short period of time. Cleaning up garbage on the street or in an alley is one example. This is not the case with infrastructure complaints. So although detritus and infrastructure complaints may have the same priming effect on deviant behavior, over a long period of time, infrastructure complaints are likely to persist for longer periods and subsequently expose more individuals to the stimulus.

A design-based reason for making this distinction is that infrastructure and detritus complaints are likely to be confounded with other correlates of crime, but each in different ways. Infrastructure complaints take more monetary or social capital to fix compared with detritus complaints. A community group could organize the cleanup of garbage in a back alley, but it would take more social capital to lobby to get a sidewalk repaired or a vacant building demolished. So, infrastructure complaints may be confounded with poverty. Detritus is correlated with commercial establishments (Taylor, Koons, Kurtz, Greene, & Perkins, 1995). So crime and detritus physical disorder may have a spurious correlation if one does not control for place-based generators of crime, such as bars (McCord, Ratcliffe, Garcia, & Taylor, 2007; Steenbeek, Völker, Flap, & Oort, 2012). In subsequent models, I nest street units within neighborhoods to control for demographic characteristics, such as poverty. I also include a wide variety of microplace crime generators (such as alcohol outlets) to control for whether a street unit is a busy place (Wilcox et al., 2004).

Study Design

311 Calls in Washington, D.C

The website, opendata.dc.gov, provides open access to individual 311 service requests along with a textual description, date, and location of the request, and further fields to determine a resolution to the request in Washington, D.C. (The opendata.dc.gov website contains all the data used in this article.) Table 1 displays the most frequent of those requests for the year 2010. For this analysis, text strings were coded into calls for service that reasonably reflect signs of physical disorder. Those 311 calls for service used in the analysis are presented in Table 1.

Table 1.

Detritus and Infrastructure 311 Calls for Service, D.C. (2010).

Detritus calls		Infrastructure calls
Sanitation enforcement	7,765	Pothole	5,681
Alley cleaning	5,870	Streetlight repair	5,565
Abandoned vehicle—On public property	5,352	Rat abatement	3,123
Yard waste—Missed	4,369	Sidewalk repair	2,330
Abandoned vehicle—On private property	2,900	Alley light repair	1,670
Illegal dumping	2,808	Graffiti removal	1,494
Recycling collection—Missed	2,604	Sign replace	1,369
Dead animal pickup	2,526	Street repair	1,362
Street cleaning	2,389	Alley repair	883
Grass and weeds mowing	554	Vacant lot	794
Litter can—Collection	341	Curb and gutter repair	546
		Sign missing	512
		Utility repair	457
Total detritus	37,478	Total infrastructure	25,786

Dependent Variable and Spatial Units of Analysis

The dependent variable used in this analysis is a set of reported crimes in Washington, D.C., in 2011. Crimes in 2011 were chosen to ensure that the effects of 311 calls (in 2010) would be exogenous.³ Those crimes are homicide, sex offense, robbery, assault with a deadly weapon, larceny, burglary, stolen auto, theft from auto, and arson. There were a total of 32,440 crimes reported in the 2011 data set (after eliminating 299 records with missing spatial coordinates or those with spatial coordinates listed outside of D.C.). These are all the crimes disseminated by D.C., and they are all modeled as broken windows theory makes no distinction between different types of crimes.⁴

The crime data publicly available are disseminated with spatial coordinates, but only at the nearest street midpoint and intersection. These will be the microplace units of analysis in this study. A street centerline file was used to create the citywide sample of street units. That is, the sample includes all street units within D.C., not just street units that had either a reported crime and/or a reported 311 call for service. The street midpoints are calculated as the middle of the line string, and as the centerline file is created for geocoding, it is conveniently split up into segments of face blocks (e.g., 100-199 Main St.). Intersections are locations where any two street centerlines overlap or touch one another. This resulted in some locations that were not well defined or irregular (e.g., traffic circles, highways, T intersections). Points within 5 m of one another were manually inspected and those that were near overlapping in the network were pruned. This, for the most part, ended up being street midpoints pruned around areas with many intersections. This resulted in a total of 21,506 street units in the study, with 8,172 (38%) being intersections.⁵

The micro-level street units are nested within neighborhoods that are approximated using a regular grid that is based on cells of size 500 m by 500 m. This is done because crime recorded at street segments and intersections has a large proportion that fall on the border of census geographies (Stucky & Ottensman, 2009). Nesting street units within neighborhoods allows one to use fixed effects to control for any neighborhood invariant characteristic (Brame, Bushway, & Paternoster, 1999) such as unmeasured levels of collective efficacy or percent in poverty. The neighborhood fixed effects are estimated using dummy variables for each neighborhood (Allison & Waterman, 2002). For the subsequent analysis, street units within grid cell neighborhood areas that do not vary in either the independent or dependent variables used need to be eliminated, as they are not identified in fixed effects models. That is, entire neighborhood grid cells need to be eliminated if they have zero crime or zero 311 calls across all nested street units, in which case, the fixed effects for the grid cell cannot be estimated. This pruning leaves a total of 20,319 street units in 553 grid cell neighborhoods.⁶ These tended to be locations with very few nested street units, such as grid cells partially containing the Anacostia and Potomac rivers. Note that spatial lags of the disorder variables were calculated before these particular grid cells were dropped, so there are no missing data with the 311 measures.

To visualize the regular grid that is as a proxy for neighborhoods, as well as the spatial distribution of crime and 311 calls for service, Figure 1 displays the aggregate measures of the mean number of crimes or calls per service for the nested street units. The map in the top left corner displays variably sized circles to show the number of street units within a particular grid cell. One can see from the sized circles in the top left that the majority of grid cells in the city have a large number of observations, 30 or more. Only a few cells (mostly those that share much of their area with waterways or parks) do not have very many nested street units. This suggests that the fixed effects analysis will not have serious problems controlling for omitted neighborhood-level effects, as each grid cell neighborhood has many nested observations.

Figure 1.

Summary measures for crime and 311 calls for service at the grid used for the fixed effects estimation.

Spatial Weights and Measures

Spatial measures used in this study (such as the spatial lags of disorder measures and calculating spatial autocorrelation) were calculated using a queens contiguity spatial weights matrix based on the Thiessen (or Voronoi) polygons of each street unit (Boots, 1986).

When calculating spatial lags to use on the right-hand side in regression equations, the counts of the total 311 calls in neighboring locations will be used. When calculating measures of residual spatial autocorrelation, global Moran’s I will be used (Anselin, 1995). The null distribution of Moran’s I is generated using a spatial permutation approach by randomly reassigning the spatial labels and recalculating the statistics for 99 simulations (Anselin, 1995).

Other Control Variables

Additional control variables were included, based on the theory of crime at small places (Eck & Weisburd, 1995; Weisburd et al., 2012). Particular attention was paid to exogenous variables that would increase the general walking around population at microplaces (Wilcox & Eck, 2011; Wilcox et al., 2004), but several variables measuring other general theories of crime were incorporated.

These control variables include the following: green sites, which include community gardens and rain collector barrels, as a measure of collective efficacy (Sampson, Raudenbush, & Earls, 1997). Both signify an investment in the community by its constituents, and community gardens facilitate informal interactions that help to build social capital (Branas et al., 2011). Sidewalk cafes and wireless hot spots were included as measures of gentrification (Papachristos, Smith, Scherer, & Fugiero, 2011). Parks, trees (that are alive), and street lights were included as measures related to visibility (Donovan & Prestemon, 2012; Farrington & Welsh, 2004; Groff & McCord, 2012). Places or institutions that are crime generators, such as bus stops and subway entrances (Levine, Wachs, & Shirazi, 1986); schools (Roncek & Lobosco, 1983); recreational areas (these mainly include basketball and tennis courts); public housing (Roncek, Bell, & Francik, 1981); university areas, hospitals, shopping malls, and libraries (Eck, Clarke, & Guerette, 2007); and places that have alcohol licenses (Block & Block, 1995). Police stations were also included, although they may be expected to be a guardian preventing crimes.

In addition to these covariates, fixed characteristics of the street units were included in the model. The area of sidewalks, roads, and whether the street unit is an intersection were included as these may relate to either the accessibility or the number of individuals exposed to crime victimization. The log of the area of the Voronoi tessellation was included in the model as a set of nonlinear restricted cubic spline basis (Harrell, 2001). This is to account for the area of the street unit, such as an exposure term, but theoretically the area may have nonlinear effects. This is because very small areas are likely competing for crime being recorded nearby, so would have lower counts of crime, and very large areas (such as on highways) are likely to have little foot traffic, so would have lower counts of crime as well.

All these measures, except for aspects of built environment, parks, recreational areas, public housing, and university areas, are disseminated at the point support. Each point measure was assigned the nearest street unit. For those measures of the built environment that are areas, the area was calculated as the area intersecting the Thiessen polygon of the street unit. The natural logarithm of these areas as square meters were modeled, and zero areas were treated as having a value of 0 (i.e., log(1)). The area of the Thiessen polygon only includes land area, area over the rivers is not included (hence some street units at bridges have zero area). All measures were gathered from DC.gov and had a release date prior to 2011 (with most being a snap shot of the current conditions sometime in 2010).

Parks, recreational areas, public housing, and university areas were calculated as a dummy variable whether the Thiessen polygon of the street unit overlapped one of these areas. Due to the street units being quite small, they mostly shared the entire area when they did overlap, providing little variation to assess differing levels of exposure for these types of areas.

Modeling Strategy

The regression models used to estimate the effect of 311 calls for service on crime, conditional on the control variables, are negative binomial regression models of the form in Equations 1 and 2. In the expected value, these models are the same as a typical Poisson generalized linear regression model, but include a dispersion term to account for the fact that variance of crime is often greater than the mean. The type of dispersion term is referred to as the NB2 model (Long & Freese, 2006), so a dispersion estimate of zero would be equal to the Poisson model.

E [Y X] = e^{β_{k} X_{k}} = μ,

(1)

V [Y X] = μ (1 + α μ) .

(2)

Four sequential models will be presented, outlined in the equations labeled as Model 1 to Model 4. Model 1 is the unconditional model that includes all the 311 calls for service terms, labeled as BW. The BW term includes all six coefficients associated with the local counts and spatial lagged counts of detritus and infrastructure 311 calls. Model 2 includes the neighborhood fixed effects, $γ N_{j}$ . Model 3 does not include the neighborhood fixed effects, but does include a series of covariates for the microplace, represented by $X ψ$ Then, finally, Model 4 includes all three sets of terms.

Model 1 : y_{i j} = B W β,

Model 2 : y_{i j} = B W β + γ N_{j},

Model 3 : y_{i j} = B W β + X Ψ,

Model 4 : y_{i j} = B W β + X Ψ + γ N_{j} .

Only the coefficients related to physical disorder are reported, along with model summary diagnostics such as the log likelihood, and the degrees of freedom. Due to the length of the list of control variables, coefficient estimates of these are not included in tables in text, but are available in the supplementary material. Models are also tested for residual spatial autocorrelation using Moran’s I. A smaller value of Moran’s I indicates that there is no residual spatial autocorrelation in the model residuals—so a smaller value is better. The full set of regression coefficients, other descriptive statistics, and bivariate correlations are available in the supplementary material for the article, as well as the data files to replicate the analysis.

Analysis

Table 2 presents the results for the four regression models. The B and the SE columns stand for the linear effect and its standard error. Model 1 presents the effects of detritus and infrastructure calls for service without including the neighborhood fixed effects or other control variables. All variables are statistically significant. The coefficients generally show the expected pattern, that the local effects are stronger than the spatial effects, but all are relatively small in absolute terms. Although the spatially lagged detritus calls shrink to practically nothing, and the effect of the first-order lag is negative (but very small), the first- and second-order neighboring values of infrastructure calls are still large relative to the first-order neighbor term. This is likely due to other omitted variables, and that infrastructure complaints are capturing other variance related to poverty or disadvantage in any particular neighborhood.

Table 2.

Negative Binomial Regression Models Predicting Crime in 2011, N = 20,319.

	Model 1					Model 2					Model 3					Model 4
	B	SE	Low	High	p	B	SE	Low	High	p	B	SE	Low	High	p	B	SE	Low	High	p
Detritus	0.034	0.005	1.026	1.045	<.01	0.039	0.004	1.032	1.048	<.01	0.022	0.004	1.013	1.030	<.01	0.015	0.004	1.008	1.023	<.01
Detritus neighbor (1)	−0.003	0.001	0.995	0.998	<.01	0.001	0.001	1.000	1.003	.04	−0.002	0.001	0.996	1.000	.01	−0.001	0.001	0.998	1.000	.13
Detritus neighbor (2)	0.005	0.001	1.003	1.007	<.01	0.007	0.001	1.005	1.009	<.01	0.007	0.001	1.005	1.009	<.01	0.004	0.001	1.002	1.005	<.01
Infrastructure	0.046	0.009	1.028	1.067	<.01	0.040	0.007	1.026	1.055	<.01	0.021	0.008	1.005	1.037	.01	0.015	0.006	1.003	1.028	.01
Infrastructure neighbor (1)	0.036	0.004	1.029	1.044	<.01	0.029	0.003	1.024	1.035	<.01	0.002	0.003	0.996	1.008	.52	−0.001	0.002	0.994	1.004	.69
Infrastructure neighbor (2)	0.019	0.002	1.016	1.023	<.01	0.013	0.002	1.010	1.017	<.01	0.014	0.002	1.010	1.017	<.01	0.004	0.002	1.001	1.008	.01
Moran’s I (Pearson residuals)	0.093				<.01	−0.007				.02	0.030				<.01	−0.028				<.01
Dispersion	3.714	0.062				2.652	0.048				2.693	0.050				2.046	0.040
Log likelihood	−29,777					−28,283					−28,443					−27,256
Degrees of freedom	20,311					19,759					20,285					19,733

Note. Coefficients for the other control variables can be obtained in the supplementary material.

Model 2, which includes the neighborhood fixed effects, shows that the effect of local detritus is not mediated. Both the spatial lags are still statistically significant at the .05 level, but the coefficients are still quite small compared with the local effect. Infrastructure complaints are mediated somewhat, but the spatial effects relative to the local effect still suggest there are omitted variables in the model, presuming that these spatially lagged effects should be zero.

Model 3 includes the other control variables that vary at the street unit level, but does not include the neighborhood fixed effects. In this model, the local effects for both detritus and infrastructure complaints are further mediated, but the second-order detritus and infrastructure complaints are still statistically significant and relatively large compared with the local effect, .021 versus .014 for the infrastructure coefficients.

Model 4 incorporates both the neighborhood fixed effects and the additional micro-level control variables. The local effects of detritus and infrastructure complaints are further mediated, but still statistically significant. The second-order spatially lagged detritus and infrastructure variables continue to be statistically significant at the .05 level, but the effect is small compared with the local coefficient. For both detritus and infrastructure complaints, the local effect is .015 and the second-order spatial effect is .004. Again, the first-order infrastructure lag fails to reach statistical significance for each set of coefficients. This suggests that there are still omitted factors in the model, but the size of the coefficient is at least smaller than the local coefficient, .015 versus .004, hinting that such bias may be small. Appendix A in the supplementary material lists additional model specifications using block groups and census tracts as the neighborhood unit, and all result in the same inferences as Model 4.⁷

For the original concern that infrastructure- and detritus-related signs of disorder might have different confounds do not bear out in Models 2 and 3, with each set of detritus and infrastructure complaints showing similar levels of mediation when including the neighborhood fixed effects or the micro-level covariates. In the final reported Model 4, to within rounding detritus and infrastructure complaints have the same estimated effect in this sample, providing evidence that they each have very similar effects on subsequent criminal behavior.⁸

Although the estimates of the local effects for both detritus and infrastructure calls for service are statistically significant for all four model specifications, the coefficients are incredibly small. This is partly a function of the large sample size used here, more than 20,000 street units. For an example, based on Model 4, if one expected five crimes, and one then added five detritus or infrastructure calls for service, the total number of crimes expected would increase to 5.4. This is a higher end estimate, as neither many street units have five crimes to begin with, nor do many street units have five detritus calls for service within a year.

To estimate an effect size in terms of counts of crime based on an actual intervention someone may attempt, the model implied reduction in crimes if all detritus and infrastructure 311 calls for service on a particular street unit were eliminated based on the local coefficient (ignoring the spatial lags) were calculated. That is, the total numbers of crimes were predicted from the model, and then the predictions were generated assuming 311 calls could be eliminated entirely at the particular street unit. These predicted crime reductions were then aggregated up to the grid cell neighborhoods. This simulates the best case scenario crime reductions implied by the model if one were to go to a particular area and attempt to clean up the area and prevent all detritus- and infrastructure-related complaints in the future.

Figure 2 displays a scatterplot showing the potential reductions in crime at the 553 grid cell neighborhoods. The most notable aspect of the plot is one outlier location, which happens to correspond to the Adam’s Morgan neighborhood. At this particular location, if one eliminated the above two hundred 311 complaints, one could theoretically reduce a total of above 400 crimes. This is clearly an outlier though, partially the result of one street unit having a very high predicted number of crimes due to being an outlier in the number of alcohol outlets on the street unit.⁹ The majority of the neighborhoods follow a linear relationship that is characterized by a 50 to 1 ratio. That is, in most neighborhoods, the model would predict one needs to reduce approximately 50 detritus or infrastructure calls for service to prevent one crime. For the majority of the city, such an intervention would be unlikely to result in appreciable crime declines, even assuming the most optimistic scenario, where one could entirely prevent these particular 311 complaints in the future.

Figure 2.

Expected reduction in crime at the grid cell neighborhood level if all 311 complaints were eliminated.

Conclusion

As a test of the broken windows theory, serious crimes in Washington, D.C., in 2011 at street segments and intersections were predicted based on detritus- and infrastructure-related 311 calls in 2010. Both detritus and infrastructure 311 calls are each found to have a positive and statistically significant effect on crime, but that effect is very small in magnitude.

This is reassuring, in that, if simple cues can prime individuals to commit crime, one would hope that its effect on behavior is only slight. Although the findings here are consistent with the broken windows thesis, they simultaneously show that reducing physical disorder is unlikely to be an efficient crime-reducing tactic. The findings are also similar to those examining the relationship between crime and disorder at larger aggregations, using either 311 data (O’Brien & Sampson, 2015) or other measures of physical and social disorder (Boggess & Maskaly, 2014). This suggests these prior works were not subject to aggregation bias or omitted variables at the street unit level.

One of the motivations for the study was to assess the utility of 311 calls for service as a measure of disorder. 311 calls have an advantage over community surveys, as they are potentially objective signs of physical disorder. They have an advantage over physical audits as they are regularly collected, and so are not a single snap shot at a particular time point. 311 calls have an advantage over both surveys and physical audits as they can cover an entire city and are regularly collected. The analysis here provides some additional evidence for the predictive validity of 311 calls for service as a measure of disorder by their ability to predict future crime. Given the small effect sizes estimated in this work, it is not surprising that prior studies using systematic social observation have not consistently found a relationship between crime and physical disorder. Using this larger administrative data set at microplaces allows one to estimate a more precise effect than audits at a smaller number of places would.

Like all observational studies, this one has limitations in making causal inferences based on the data at hand. It is always the case that there could be omitted variables in the regression equations that result in physical disorder and crime having a spurious relationship. One is simply a measure of the people in the area. For example, St. Jean (2008) in interviews with offenders asks about the relationship between physical disorder and how they choose spots to sell drugs. The dealers often cite that garbage in an area is simply a reflection of the number of people conducting routine behavior, such as coming or going from a bus stop or a fast food restaurant, and makes no difference in the original decision to choose a location. To control for these routine activity hot spots, the models presented include a set of crime generator locations that are likely to increase the walking around population in the model, such as bus stops, shopping malls, and so forth. It could be these controls are insufficient though, and given the effect sizes observed are quite small, it would only take a small misspecification to result in null effects for both detritus and infrastructure calls for service.

Other potential omitted variables are those related to measures of informal social control. The original broken windows theory stated that disorder leads to fear of crime, which in turns reduces informal social control in neighborhoods. There may be other mechanisms that increase or decrease informal social control though, such as the collective efficacy of the neighborhood. These omitted variables may mediate the relationship between physical disorder and crime, because the same mechanisms that cause neighbors to intervene in crime incidents might also cause them to fix broken windows. If collective efficacy only varies at the neighborhood level, it will be controlled for in the fixed effects model specification, but if there are more micro-level variations from street to street, the neighborhood fixed effects will not be sufficient. It is also possible that only individuals vested in the community are the ones most likely to report complaints via the 311 hotline, and as such measures of disorder are confounded with collective efficacy. The opposite findings, that 311 calls have a positive association with crime, suggest this not to be the case, but such a countervailing influence may result in a decrease of the estimated effects. These limitations should be weighed in relation though to biases in self-reported measures, as well as the cost in collecting physical audits.

Although 311 systems may be an imperfect measure of disorder in a community, they are likely as good a source as any for police departments to identify problem areas of the city for community police officers to focus attention, and provide a much simpler, cheaper, and up-to-date resource than community surveys or physical audits. But, although there may be other less tangible benefits to reducing disorder in a community, such as reducing fear of crime or improving the quality of life for residents (Chappell, Monk-Turner, & Payne, 2010), these findings suggest reducing physical disorder is not viable as a crime control tactic.

Footnotes

Acknowledgements

I thank Dae-Young Kim and Marc Swatt for reviewing drafts of the article. This work is a derivative of my dissertation, and so I also thank my dissertation committee, Rob Worden, Shawn Bushway, Glenn Deane, Colin Loftin, and Graeme Newman.

Author’s Note

Data and code to replicate the analysis can be downloaded from

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Supplemental Material

Supplemental material for this article is available online.

Author Biography

Andrew P. Wheeler is an assistant professor of criminology at the University of Texas at Dallas in the School of Economic, Political, and Policy Sciences. His research focuses on the spatial analysis of crime at microplaces and practical problems faced by crime analysts. His recent work has been published in the Journal of Quantitative Criminology, Cartography and Geographic Information Sciences, the International Journal of Police Science and Management, Security Journal, and the Journal of Investigative Psychology and Offender Profiling.

References

Allison

Waterman

(2002). Fixed-effects negative binomial regression models. Sociological Methodology, 32, 247-265.

Anselin

(1995). Local indicators of spatial association—LISA. Geographical Analysis, 27, 93-115.

Armstrong

Katz

(2009). Further evidence on the discriminant validity of perceptual incivilities measures. Justice Quarterly, 27, 280-304.

Bittner

(1967). The police on skid-row: A study of peace keeping. American Sociological Review, 32, 699-715.

Block

(1995). Space, place, and crime: Hot spot areas and hot places of liquor-related crime. Crime Prevention Studies, 4, 145-184.

Boggess

Maskaly

(2014). The spatial context of the disorder-crime relationship in a study of Reno neighborhoods. Social Science Research, 43, 168-183.

Boots

B. N.

(1986). Voronoi (Thiessen) polygons. Norwich, UK: Geo Books.

Brame

Bushway

Paternoster

(1999). On the use of panel research designs and random effects models to investigate static and dynamic theories of criminal offending. Criminology, 37, 599-642.

Branas

Cheney

MacDonald

Tam

Jackson

Ten Have

(2011). A difference-in-differences analysis of health, safety, and greening vacant urban space. American Journal of Epidemiology, 174, 1296-1306.

10.

Chappell

Monk-Turner

Payne

(2010). Broken windows or window breakers: The influence of physical and social disorder on quality of life. Justice Quarterly, 28, 522-540.

11.

Community Oriented Policing Services. (2007). 311 for non-emergencies: Helping communities one call at a time (COPS Fact Sheet). Washington, DC: Office of Community Oriented Policing Services, U.S. Department of Justice. Retrieved from http://www.cops.usdoj.gov/html/cd_rom/inaction1/pubs/311NonEmergencies.pdf

12.

Donovan

Prestemon

(2012). The effect of trees on crime in Portland, Oregon. Environment & Behavior, 44, 3-30.

13.

Eck

Clarke

Guerette

(2007). Risky facilities: Crime concentration in homogenous sets of establishments and facilities. Crime Prevention Studies, 21, 225-264.

14.

Eck

Weisburd

(1995). Crime places in crime theory. Crime Prevention Studies, 4, 1-33.

15.

Executive Office of the President. (2014). Big data: Seizing opportunities, preserving values. Retrieved from http://www.whitehouse.gov/sites/default/files/docs/big_data_privacy_report_may_1_2014.pdf

16.

Fagan

Davies

(2000). Street stops and broken windows: Terry, race, and disorder in New York City. Fordham Urban Law Journal, 28, 457-504.

17.

Farrington

Welsh

(2004). Measuring the effects of improved street lighting on crime: A reply to Dr. Marchant. The British Journal of Criminology, 44, 448-467.

18.

Gau

Pratt

(2008). Broken windows or window dressing? Citizens’ (in)ability to tell the difference between disorder and crime. Criminology & Public Policy, 7, 163-194.

19.

Groff

McCord

(2012). The role of neighborhood parks as crime generators. Security Journal, 25, 1-24.

20.

Harcourt

Ludwig

. (2006). Broken windows: New evidence from New York City and a five-city social experiment. The University of Chicago Law Review, 73, 271-320.

21.

Harrell

(2001). Regression modeling strategies: With applications to linear models, logistic regression, and survival analysis. New York, NY: Springer-Verlag.

22.

Hinkle

(2013). The relationship between disorder, perceived risk, and collective efficacy: A look into the indirect pathways of the broken windows thesis. Criminal Justice Studies, 26, 408-432.

23.

Hipp

(2010). Resident perceptions of crime and disorder: How much is “bias,” and how much is social environment differences? Criminology, 48, 475-508.

24.

Kahneman

(2011). Thinking fast and slow. New York, NY: Macmillan.

25.

Klinger

Bridges

(1997). Measurement error in calls-for-service as an indicator of crime. Criminology, 35, 705-726.

26.

Levine

Wachs

Shirazi

(1986). Crime at bus stops: A study of environmental factors. Journal of Architectural and Planning Research, 3, 339-361.

27.

Link

Kelly

Pitts

Waltman-Spreha

Taylor

(2014). Reversing broken windows: Evidence of lagged, multilevel impacts of risk perceptions on perceptions of incivility. Crime & Delinquency, 63, 659-682. doi:10.1177/0011128714555606

28.

Long

Freese

(2006). Regression models for categorical dependent variables using Stata. College Station, TX: Stata Press.

29.

McCord

Ratcliffe

Garcia

Taylor

(2007). Nonresidential crime attractors and generators elevate perceived neighborhood crime and incivilities. Journal of Research in Crime and Delinquency, 44, 295-320.

30.

O’Brien

(2015). Custodians and custodianship in urban neighborhoods: A methodology using reports of public issues received by a city’s 311 hotline. Environment & Behavior, 47, 304-327.

31.

O’Brien

Sampson

(2015). Public and private spheres of neighborhood disorder: Assessing pathways to violence using large-scale digital records. Journal of Research in Crime and Delinquency, 52, 486-510.

32.

O’Brien

Sampson

Winship

(2015). Ecometrics in the age of big data: Measuring and assessing “Broken windows” using large-scale administrative records. Sociological Methodology, 45, 101-147.

33.

O’Brien

Winship

(2017). The gains of greater granularity: The presence and persistence of problem properties in urban neighborhoods. Journal of Quantitative Criminology. Advance online publication, 33, 649-674.

34.

Papachristos

Smith

Scherer

Fugiero

(2011). More coffee, less crime? The relationship between gentrification and neighborhood crime rates in Chicago, 1991 to 2005. City & Community, 10, 215-240.

35.

Raudenbush

Sampson

(1999). Ecometrics: Toward a science of assessing ecological settings, with application to the systematic social observation of neighborhoods. Sociological Methodology, 29, 1-41.

36.

Rogers

(1999). 311 non-emergency. Law Enforcement Technology, 26, 68-70.

37.

Roncek

Bell

Francik

(1981). Housing projects and crime: Testing a proximity hypothesis. Social Problems, 29, 151-166.

38.

Roncek

Lobosco

(1983). Effect of high schools on crime in their neighborhoods. Social Science Quarterly, 64, 598-613.

39.

Rosenfeld

Fornango

Rengifo

(2007). The impact of order-maintenance policing on New York City homicide and robbery rates: 1988-2001. Criminology, 45, 355-384.

40.

St. Jean

. (2008). Pockets of crime: Broken windows, collective efficacy, and the criminal point of view. Chicago, IL: The University of Chicago Press.

41.

Sampson

Raudenbush

(1999). Systematic social observation of public spaces: A new look at disorder in urban neighborhoods. American Journal of Sociology, 105, 603-651.

42.

Sampson

Raudenbush

(2004). Seeing disorder: Neighborhood stigma and the social construction of “Broken windows.” Social Psychology Quarterly, 67, 319-342.

43.

Sampson

Raudenbush

Earls

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918-924.

44.

Steenbeek

Völker

Flap

Oort

(2012). Local businesses as attractors or preventers of neighborhood disorder. Journal of Research in Crime and Delinquency, 49, 213-248.

45.

Stucky

Ottensman

(2009). Land use and violent crime. Criminology, 47, 1223-1264.

46.

Swatt

Varano

Uchida

Solomon

(2013). Fear of crime, incivilities, and collective efficacy in four Miami neighborhoods. Journal of Criminal Justice, 41, 1-11.

47.

Taylor

(2001). Breaking away from broken windows: Baltimore neighborhoods and the nationwide fight against crime, grime, fear, and decline. Boulder, CO: Westview Press.

48.

Taylor

Koons

Kurtz

Greene

Perkins

(1995). Street blocks with more nonresidential land use have more physical deterioration. Urban Affairs Review, 31, 120-136.

49.

Wallace

(2015). A test of the routine activities and neighborhood attachment explanations for bias in disorder perceptions. Crime & Delinquency, 61, 587-609.

50.

Wallace

Hedberg

Katz

(2012). The impact of foreclosures on neighborhood disorder before and during the housing crisis: Testing the spiral of decay. Social Science Quarterly, 93, 625-647.

51.

Wallace

Louton

Fornango

(2015). Do you see what I see? Perceptual variation in reporting the presence of disorder cues. Social Science Research, 51, 247-261.

52.

Weisburd

Groff

Yang

(2012). The criminology of place: Street segments and our understanding of the crime problem. New York, NY: Oxford University Press.

53.

Weisburd

Groff

Yang

(2014). Understanding and controlling hot spots of crime: The importance of formal and informal social controls. Prevention Science, 15, 31-43.

54.

Wilcox

Eck

(2011). Criminology of the unpopular. Criminology & Public Policy, 10, 473-482.

55.

Wilcox

Quisenberry

Cabrera

Jones

(2004). Busy places and broken windows? Toward defining the role of physical structure and process in community crime models. Sociological Quarterly, 45, 185-207.

56.

Wilson

Kelling

(1982). The police and neighborhood safety: Broken windows. The Atlantic, 127, 29-38.

57.

Wilson

Kelling

(1989). Making neighborhoods safe. The Atlantic, 263, 46-52.

58.

Worrall

(2006). The discriminant validity of perceptual incivility measures. Justice Quarterly, 23, 360-383.

59.

Yang

(2010). Assessing the spatial-temporal relationship between disorder and violence. Journal of Quantitative Criminology, 26, 139-163.

60.

Yang

Pao

(2015). Do we “see” the same thing? An experimental look into the black box of disorder perception. Journal of Research in Crime and Delinquency, 52, 534-566.

61.

Zimbardo

(1973). A field experiment in auto shaping. In Ward

(Ed.), Vandalism (pp. 85-90). London, England: The Architectural Press.

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

0.13 MB