Introduction

Review of the Literature

Homicide clearance has been a focus of study by criminologists for decades. Riedel and Boulahanis (2007) examined the confusion around the precipitous drop in clearance rates that occurred from the 1960s, where about 90% of all cases were cleared by arrest, which then plummeted to about 65% by 2005 and more recently reported by the FBI UCR as 62% in 2017 (FBI, 2017).

According to Riedel and Boulahanis (2007) homicide clearance involves the following:

There are a number of factors that can influence a jurisdiction’s clearance rates. In disentangling these findings, Keel et al. (2009) set out to explore why clearance rates stay low. They examined this problem using five different criteria:

1. Management and resources. These included the amount of personnel and administrative support provided to investigators and prosecutors in pursuing a case;

2. Investigative procedures. These involved the extent to which detectives are able to obtain information from witnesses and victims to help in the investigation;

3. Analytical processes. These pertained to the degree to which investigators had access to technology and analytical techniques that helped in identifying offenders;

4. Contextual and demographic factors. These relate both to the characteristics of the communities in which the homicides were taking place and to the size of the police jurisdictions affected;

5. Political influences. The importance of media accounts and political concerns about rising crime and how that relates to the extent to which activities by police, prosecutors and other judicial authorities are influenced in the priority they give to clearing homicide cases.

In a recent analysis of an intervention administered in the Boston Police Department, Braga and Dusseault (2018) report gains in clearance rates as a result of added resources for homicide detectives. Further, in a survey of 55 agencies, Keel et al. (2009) found support for the idea that an increase in management resources, improved investigative procedures, and application of advanced analytical improvements could improve clearance rates, a finding which was also reported by Wellford et al. (2019) and Cook et al. (2019). But, importantly, they report the offsetting salience that contextual and demographic factors had in suppressing reporting by victims and distrust that kept victims from providing testimony to assist in arrests.

Maguire et al. (2010) explored the competing effects of circumstances and administrative practices on clearance rates. Interestingly, they found that interagency cooperation impeded case clearances as homicide cases began to increase; but, with the growth in homicide occurrence, a lack of interagency cooperation can also impede clearance. This research was set in the context of the rise of gang related violence and an increased reluctance of victims to come forward in support of adjudication of the cases, which they identified as impediments to clearance. In other research, comparing US trends with Canada, Regoeczi et al. (2000) showed the importance of a number of factors; for the US this included gender of victim, age and circumstances around the offense, where only the latter influence clearance rates in Canada. The relevance of circumstances surrounding the offenses, as well as the characteristics of the victims and offenders influenced US rates. These findings were also supported by Jarvis and Regoeczi’s (2009) research, as they found that the relevance of circumstances surrounding the offenses and the characteristics of the victims and offenders also impact homicide clearance rates in the US. Further, in a recent paper, Regoeczi et al. (2020) examined a national sample of homicides and concluded that social context, including the effects of different social statuses of victims, can be an important factor relevant to clearance outcomes. Their study led them to suggest that research on clearance rates could benefit from more community level inquiries, including the effects that differences across economically diverse neighborhoods have on these inquiries.

Additionally, research conducted by Petersen (2016, 2017) has begun to examine homicide clearances by evaluating two separate criteria outlined previously by Keel et al. (2009). In his most recent piece, Petersen examined how important contextual and demographic factors at the neighborhood level (for Los Angeles County, California) may influence homicide clearance rates. In sum, Petersen (2017) found that although there are multiple non-racial variables that influence homicide clearances, homicides that occur in predominately minority areas remain less likely to be closed. Importantly, his research not only included key individual-level covariates, he also incorporated multiple neighborhood-level covariates and agency-level covariates. Prior to conducting this research, Petersen (2016) also evaluated how neighborhood factors might influence the media’s coverage of homicide cases throughout Los Angeles County, California. In this study, Petersen (2016) found “that the level of economic disadvantage and percentage of minority residents in/around the crime scene neighborhood negatively affect the presence/absence and rate of newspaper coverage” (p. 25).

Along these lines, Brunson and Wade (2019) report on willingness to testify about homicides from interviews of 50 young Black men, who were residents of high-crime neighborhoods in Brooklyn and the Bronx. They found that these individuals had considerable knowledge about illegal gun markets and the resulting bloodshed. They concluded that, “distressed milieus reliably fail to produce cooperative witnesses as a result of the cumulative impact of anti-snitching edicts, fear of retaliation, legal cynicism, and high-risk victims’ normative views toward self-help” (Brunson & Wade, 2019, p. 623). When Brunson and Wade (2019) discuss “distressed milieu” they are referring to locations in which there is sufficient social disorganization to make the concerns about being vulnerable to retaliation and reprisal real enough to compel potential witnesses to avoid testifying. Like the other research reported above this work has concentrated on the characteristics of victims and offenders when looking at what defines the circumstances that influence homicide outcomes. While social disorder may create these conditions of non-compliance, there is little known about how this manifests itself in terms of the spatial allocation of risky features in the environment that influence clearance rates. For example, is the presence of environmental features such as bars or section VIII housing contributing factors to this sense of vulnerability? If so, does this vulnerability spatially manifest across locations? Knowing how these risk factors configure to create conditions that frame the difficulties that police face in collecting facts and statements about homicide cases would assist in clearing them.

For example, Griffiths and Tita’s (2009) study sought to explain why public housing developments in Los Angeles were plagued with considerably higher violent crime rates than areas that were similar across multiple socio-demographic variables. As such, they attempted to determine whether or not housing developments could be categorized as “hotbeds,” “magnets,” and/or “generators” of violent crimes. After reviewing 20 years of homicide data from the Southeast Policing Area of Los Angeles, they not only characterized these developments as “hotbeds” of violent crime but they also concluded that a majority of these violent crimes were committed by offenders who resided within said developments (Griffiths & Tita, 2009).

Understanding the collective influences of environmental features that attract crime and generate illegal behaviors has proven valuable for policing (Brantingham & Brantingham, 1995; Brantingham et al., 2020; Braga et al., 2019; Kennedy et al., 2018). The Law of Crime Concentration (Weisburd, 2015), Crime Pattern Theory (Brantingham & Brantingham, 1981), Theory of Risky Places (Kennedy et al., 2016, 2018), and other theoretical frameworks within the domain of criminology and criminal justice (Brantingham & Brantingham, 1981, 1995; Park et al., 1925; Quetelet, 1984; Shaw & McKay, 1969; Wikström, 2010). Brantingham and Brantingham (1995) explain that spatial crime patterns, and their stability over time, are a function of the “environmental backcloth” of the area under study. This backcloth is dotted with “crime attractors” and “crime generators.” Attractors include features of the environment that entice offenders to come to places to commit crime. Generators are represented by increased opportunities for crime that emerge from the collection of more people into areas following specific types of behavior, simply because of the increased volume of interaction taking place in these areas. Certain features of the landscape exert spatial influences on human behaviors that can affect a place’s vulnerability to crime, which is why crimes emerge, cluster and persist over time (Brantingham & Brantingham, 1995; Brantingham et al., 2020; Caplan & Kennedy, 2016; Kennedy et al., 2016; National Academies of Sciences, Engineering, and Medicine, 2017). Braga and Clarke (2014), Kennedy et al. (2018), Barnum et al. (2017), and others (Braga & Weisburd, 2010) present compelling evidence to focus on certain types of environmental features at chronically crime-prone areas because these features increase the probability crime will occur by attracting offenders, enabling illegal behavior, and confounding agents of social control in their efforts to contain or suppress their negative outcomes. Also, the work of Simon (1991) and Leovy (2015), in their studies on homicide investigation, support examining features of the built environment and their influence on case processing.

Environmental theories and related research provide some insights into how individual persons select and use the environments they occupy and the impact this has on crime outcomes. The potential for varied homicide clearance rates across different geographies may also rest on similar notions that these crimes occur where characteristics about the places differentially affect their solvability. Perhaps where some homicides occur there also exists particular features of the landscape that create unique settings that limit or foster police investigations. This reasoning further accommodates the ideas of situational crime prevention, Rational Choice Theory, and Opportunity Theory (Clarke, 1997; Clarke & Eck, 2005; Guerette & Bowers, 2009; Hunter & Jeffery, 1997) as it relates to a motivated offender’s selection of certain locations for homicidal acts (Cohen & Felson, 1979; Cohen et al., 1981; Groff & La Vigne, 2002). Particular aspects of the spatial contexts of places could be perceived as raising the risk that homicides will be solved easier, perhaps due to lines-of-sight or witness participation, for example, that could make some areas less suitable locations for homicides by offenders who do not want to be caught.

The limitations in managerial resources committed to clearing homicide cases combined with efforts to enlist witnesses and victims to participate in prosecuting cases creates real problems for investigators and judicial officials. But, in most research on this topic, attention has been focused on cross-jurisdiction comparisons operating on the assumption that both administrative and contextual factors can be used to explain outcomes, without understanding their relative importance. To disentangle the effects of these two explanations, we set out to control for administrative procedures and political commitment through the examination of one police agency, the New York City Police Department (NYPD), which operates in a wide variety of communities and throughout five boroughs. Applying this control on inter-agency differences, assuming an across department commitment to pursue homicide cases in an evenhanded way, allows us to explore in detail the impacts that contextual factors have on influencing clearance rates at micro places throughout the jurisdiction. These contextual factors can include the demographic and socio-economic characteristics of neighborhoods, in addition to the varying impacts that the different environmental conditions across the city might have in affecting these results. The combined effects of some features or qualities of the landscape, such as, drug markets, public housing, and so on could influence crime reporting or the suppression of information available to police in leading to an arrest.

Study Setting, Data and Methods

Collectively, scholars have concluded that crime, at the national level, decreased dramatically during the 1990s (Blumstein & Wallman, 2006; Levitt, 2004; Walker, 2015; Weisburd et al., 2014). Despite this acknowledgment, a contentious debate surrounding the reasons for that decline and whether some cities experienced a greater decrease than others still continues. For example, some researchers suggest the crime decline experienced throughout NYC was unparalleled (Weisburd et al., 2014; Zimring, 2007), especially when examining the 20-year time period from 1990 to 2010 (Weisburd et al., 2014). However, other researchers have suggested the crime decline experienced in NYC was not completely unique when compared to other major American cities and the conditions that allowed the decline to occur are far more complex than originally thought (see Baumer & Wolff, 2014; Bowling, 1999; Fagan et al., 1998; Levitt, 2004). This continued debate caused NYC to become an epicenter of both policing and crime-based research.

Although there is a sizable body of research that examines these issues as they pertain to the crime decline experienced throughout NYC, this review focuses on those studies that examined homicides, as this study looks to examine open versus closed homicide cases. Fagan et al. (1998) began by examining the homicide trend for the city as it compares to other cities throughout the U.S. and then examined the homicide trend for NYC by borough for the years 1990 through 1995. In agreement with other research, Fagan et al. (1998) concluded the homicide decline for that period of time was not unprecedented. That being said however, their research found that new patterns emerge when homicide data is disaggregated by weapon (i.e., gun), gender and age. For example, Fagan et al. (1998, p. 1289) found that homicides committed with a firearm initially increased before falling back to previously levels whereas non-firearm homicides trended downward from start to finish.

Messner et al. (2007) examined the possible effects that both broken windows policing and the cocaine market had on homicides across NYC police precincts from 1990 to 1999. They concluded that “the effects of misdemeanor arrests and cocaine prevalence emerge for gun-related but not for non-gun-related homicides” (Messner et al., 2007, p. 386). Similarly, Chauhan and Kois (2012) examined the effects that misdemeanor enforcement and drug markets have on gun-related homicides at the precinct level in NYC from 1990 through 1999. Two very important findings emerge from this study. First, their analyses revealed, “one quarter of NYPD precincts were responsible for driving the overall decrease in homicide rates” (Chauhan & Kois, 2012, p. 20). Second, and akin to previous work, their final results were mixed, thus alluding to the fact that no single aspect examined could be linked inextricably to the homicide decline of the 1990s (Chauhan & Kois, 2012).

Furthermore, there were two additional studies that also examined the decline in homicides in NYC from 1988 through 2001. This first study, conducted by Rosenfeld et al. (2007), critically examined the impact of order-maintenance policing (via proactive policing of quality of life offenses) on both homicides and robberies. Their research found that in the precincts where arrests for quality of life offenses increased, both robberies and homicides decreased (Rosenfeld et al., 2007, p. 366). Additionally, they also reported that as the level of disorder decreased in each precinct, so too did the amount of homicides and robberies (Rosenfeld et al., 2007, p. 367). Interestingly, these findings differ from Greenberg’s (2014) study, as he only focused on examining the effects that misdemeanor arrests had on homicides, robberies and aggravated assaults. In sum, he found “no evidence that misdemeanor arrests reduced levels of homicide, robbery, or aggravated assaults” (Greenberg, 2014, p. 154) and suggested that the decline experienced for these felony crimes must have been a result of the intersection of other factors.

In agreement with the key findings from this review, Karmen (2000) believes that the 1990s crime drop was the result of the intersection of a multitude of factors during that time period. For example, Karmen (2000) posits that some of the following conditions could have played a pivotal role in the crime decline: (1) the recovery of the economy; (2) the change in the number of students that not only finished high school but also entered college; (3) the weaning crack epidemic and the de-escalation of the arms race associated with the drug market; (4) the changes in the police department, which ranged from the technological improvements, the overall reengineering of the department, and the switch to more proactive policing styles; (5) the increase in the prison population; and even (6) the residual impact of the AIDS epidemic (pp. 257–258). Despite the mixed findings of the collective research, there is one aspect that remains constant—from 1990 through 1998, homicides dropped 72% (i.e., from 2,245 to 633) in NYC (Karmen, 2000, p. 25).

The NYPD employs just over 35,000 uniformed officers (NYPD, 2016) with approximately 5,700 detectives assigned to the Detective Bureau (NYPD, 2018). The Detective Bureau is organized across the eight patrol boroughs (i.e., Manhattan North, Manhattan South, the Bronx, Brooklyn North, Brooklyn South, Queens North, Queens South and Staten Island), each of which have their own homicide squad (NYPD, 2018). On average there are anywhere between 10 and 20 detectives assigned to each, with the exception of Brooklyn North and the Bronx, where the averages are slightly higher and may range between 20 and 30 investigators. Typically, homicide detectives remain assigned to cases in their respective boroughs and only under the most extreme circumstances would they be temporarily allowed to aid in homicide cases in another patrol borough.

In the early 1990s, Commissioner Bratton restructured the NYPD in an attempt to decentralize the agency (Nagy & Podolny, 2008). He aimed to grant the Commanding Officers more autonomy within their commands (that is, so that they could deal with their individual crime issues and constituencies in a manner that suited their distinct needs) while also allowing the upper command staff the ability to maintain control over the decisions and actions of commanding officers (Nagy & Podolny, 2008). For our purposes here, it is noteworthy that the general administrative rules were common across the department and differences among precincts did not rise to a level that would create significant or distinct differences in intentions to clear homicide cases or the resources devoted to achieving these outcomes for every reported homicide.

Homicide incident clearance data was obtained from the Washington Post. In 2018, the news agency ran a story that mapped more than 52,000 homicides in major American cities across the United States (Lowery et al., 2018). New York City provided 2 years of data to the Washington Post, from 2016 through 2017, which we used for this study. These geo-located data, including the incident location, whether an arrest was made, and basic demographic information about each victim, were made publicly available online via GitHub in comma-delimited format (https://github.com/washingtonpost/data-homicides). Consistent with the FBI Uniform Crime Reporting Program, homicides were operationalized as murder and non-negligent manslaughter but excluded suicides, accidents, justifiable homicides and deaths caused by negligence. Homicides were closed by arrest when police reported that to be the case. Cases were counted as closed without arrest if they were reported by police to be “exceptionally cleared”—whereby there was sufficient evidence to make an arrest but an arrest was not possible (e.g., the suspect died). All other incidents were classified as having no arrest, thus remaining open or uncleared. Homicides are considered “cleared” if they were closed with an arrest or by the above mentioned, “exceptionally cleared.” Based on the Washington Post data, the national clearance rate for homicide was determined to be 49%; the clearance rate for homicides in New York City was 64%. Table 1 shows counts of these data for the NYC Boroughs examined in this study. Figure 1 presents hot spot maps of cleared and uncleared homicide cases in New York City. While there are several spatial overlaps, it is also evident that distinct areas of the city exhibit clusters of only cleared or uncleared cases.

Table 1. Descriptive Information for Homicides in New York City.

View larger version

Figure 1. Distinct areas of New York City exhibit clusters of cleared or uncleared homicide cases.

Potential environmental factors that could influence homicides and related clearance rates at particular places within New York City’s environmental backcloth were obtained from NYPD’s National Institute of Justice “Policing by Place” project (award #2013-IJ-CX-0053), archived with the National Archive of Criminal Justice Data (NACJD). These data measures were initially informed by professional insights provided by the NYPD and related spatial datasets were originally obtained by the NYPD as shapefiles compiled from numerous local government agencies: the Department of Consumer Affairs, Department of Financial Services, Department of City Planning, Department of Environmental Conservation, Department of Information Technology and Telecommunications, Department of Parks and Recreation, the New York City Housing Authority, and the New York State Liquor Authority. As recommended by Caplan and Kennedy (2016) these data were then ground-truthed for accuracy and checked for construct and content validity by the NIJ research team. In total, we obtained 45 environmental features of the New York City landscape as potential risk factors (see Appendix 1 for the complete list).

RTM was used to conduct spatial analyses of the relationships between cleared and uncleared homicide cases and environmental factors of the NYC landscape. RTM offers an evidence-based and statistically valid way to diagnose spatial relationships among datasets, and to identify locations where the likelihood of particular outcomes will be high (Kennedy et al., 2011). Detailed instructions for conducting RTM are available in the extant literature (Caplan & Kennedy, 2016; Caplan et al., 2014). Risk terrain models were produced using RTMDx software,1 which has been used for similar purposes in several research studies across multiple jurisdictions. Caplan and Kennedy (2016) provide details about the RTM process and statistical methods performed by RTMDx, which involve Bayesian probabilities, cross-validations, and Poisson and negative binomial regressions. RTMDx outputs are tabular and cartographic; for each significant risk factor, tabular outputs include a relative risk value (RRV), which is the exponentiated factor coefficient (i.e., relative weight), and the optimal operationalization and distal extent of spatial influence. A risk terrain map is also produced with relative risk scores (RRSs) assigned to each micro place to convey the full range of relative spatial risks of outcome events (i.e., un-cleared homicides) throughout the study area.

RTM is a key tool in risk based policing (Kennedy et al., 2018). As Kennedy et al. (2018) explain, RTM extends the investigation from the crime incidents to the spatial contexts in which these incidents emerge or persist, offering the analytical assessment needed to inform police decision-making. RTM provides the framework for diagnosing these factors in a way that helps us understand the circumstances under which crime occurs. Risk terrain maps articulate micro-level places where conditions are suitable for illegal behavior and most likely for crimes to occur. Further, RTM has been shown to articulate officers’ “gut” feelings and perceptions of risk at places beyond merely referencing past occurrences of reported crimes. RTM provides an indication of what risk factors might be at the root of crime problems and, thus, helps police devise a problem solution. In this way, RTM helps to make information that comes from complaints about problem areas actionable and relevant to service delivery and public safety and its outputs would be particularly useful in sorting out the factors that contribute to homicides. It should, as a consequence, be equally insightful in providing clues to assist in clearing these homicides when they occur.

As pointed out above, although NYPD is a single organization that serves the entire City of New York, administrative and command distinctions can exist across boroughs based on local contexts. A citywide RTM analysis would generalize such distinctions, but could miss nuances of the micro settings within each borough where homicides could occur. For this reason, separate risk terrain models were run for each borough, using the same analysis parameters for each. We specified a cell size of 200 feet and a block length of 400 feet (the average block length in NYC) as units of analysis for RTM (Caplan et al., 2013). Prior empirical research by Taylor and Harrell (1996) and Taylor (1997) suggests that “behavior settings” are crime-prone places that typically comprise just a few street blocks (Taylor, 1988). Groff and Lockwood (2014) show that the spatial influences of environmental features located in these settings extends no further than just a few blocks and decays with distance. Based on these insights, we decided not to evaluate spatial influences beyond three blocks for our study.

Additional parameters for the RTM amalyses included operationalization, maximum spatial influence, and analysis increments. Operationalization refers to how the spatial influence of each environmental feature will be tested. Caplan (2011) explains that the spatial influence of environmental features may be operationalized as proximity (i.e., being near a feature increases risk) or density (i.e., a cluster of features increases risk). RTMDx can test both operationalizations and empirically select the most appropriate one. Maximum spatial influence defines the geographic extent to which environmental features’ influences on crime extends (i.e., the influence of bars may extend to one, two or three blocks). Because research has found that the spatial influence of features typically extends within just a few blocks, we tested the spatial influence of each environmental feature to a maximum extent of three blocks. Finally, analysis increments refer to the level of detail at which spatial influence is assessed (i.e., half-block or whole-block increments). Appendix 1 displays this parameter for each feature tested.

Results of the RTM analyses for open and closed homicide cases in each Borough (Manhattan, Brooklyn, the Bronx and Queens) are displayed in Table 2. Maps, such as Figure 2 depicting the combined spatial influences of significant risk factors and the highest risk places within each model, were exported from RTMDx software as ArcGIS shapefiles. RRS for each micro-place in the risk terrain map was then standardized as a Z-score (for comparisons across all eight models) and used as the independent variable for subsequent regression analyses, discussed below.

Table 2. Risk Terrain Modeling Findings for Open and Closed Homicides.

View larger version

Figure 2. Risk terrain maps of homicide cases in Queens borough.

Following the identification of significant risk factors, we measured whether the RRS for open and closed cases significantly predicts the status of homicide incidents. We explored this research question through a logistic regression model with the 590 homicides occurring during the study period as the unit of analysis. The dependent variable was a binary measure with solved cases coded as “1” and open cases “0.” The models have two independent variables of interest. The first is a standardized measure of the RRS for closed homicides. The second is a standardized measure of the RRS for open homicides. For both independent variables, RRS values were standardized according to the range of cell values throughout the surrounding borough. Each homicide was assigned the value of its encompassing cell. If RRS values are truly predictive of homicide status, then we would expect the RRS for closed cases to be significantly related to increased likelihood of case closure and the RRS for open cases to be significantly related to decreased likelihood of case closure. The effect of both RRS variables on homicide clearance were reported as Odds Ratios (ORs).

To control for the effect of neighborhood factors on homicide investigations, we include a concentrated disadvantage index as a control variable. This variable was an index of the standardized values of the following measures, which were all collected at the precinct level from the Infoshare Online website (http://www.infoshare.org/main/public.aspx): percentage of families receiving food stamps, percent black or Hispanic residents, percentage of families below the poverty line, percentage of single headed female households with children, and median household income. All measures were 5-year averages (2012–2016). Standard errors for all model covariates were calculated across each of the 722 police precincts in our study setting to control for any unobserved precinct-level effects on homicide clearance (e.g., number of detectives, investigative strategy, etc.).

Findings are presented in Table 3. Both the RRS for closed and open cases were statistically significant with effect sizes in the expected direction. For every 1-unit increase in the standardized RRS for closed cases the likelihood that a homicide would be solved increased by 25% (OR = 1.25; p = .04). For every 1-unit increase in the standardized RRS for open cases, the likelihood that a homicide would be solved decreased by 20% (OR = 0.80; p < .01). These findings show that spatial risk factors, as identified through RTM, may influence the solvability of homicide events in New York City.

Table 3. These Findings Show that Spatial Risk Factors, as Identified Through RTM, May Influence the Solvability of Homicide Events in New York City.

View larger version

Discussion and Conclusions

1.
From the Rutgers University Center on Public Security: www.rutgerscps.org/software.html

2.
Currently, there are 77 precincts in NYC. However, the Staten Island precincts were removed from this analysis to minimize the possibility of confounding factors affecting this analysis. For example, SI is the only borough that does not have transit districts and police service areas, which may limit the resources that could potentially support each precinct detective squad in investigating the homicide. Second, in terms of size, there are only four precincts in SI, when there are 12 in the Bronx, 22 in Manhattan, 23 in Brooklyn, and 16 in Queens. Third, in studying NYC crime trends, this was also done in previous research (see Fagan et al., 1998). Lastly, the Central Park precinct was removed from the analysis because it is the only non-residential precinct in NYC. Thus, there are only 72 precincts in the final analysis.