Identify a true statement about the relationship between intelligence and crime and delinquency.

Summary

Crime rates vary. In some populations and in some periods, the prevalence of crime is much greater than in other populations and at other time periods. Accounting for these findings is an enormously important task because if we can understand the causal processes that underlie variation then we may be in a position to enact policy changes that can bring about changes in the volume of crime in society at any given point in time. Criminologists have made significant progress in both identifying the “risk factors” that seem to relate to higher rates of offending and in constructing plausible explanatory theories that can potentially account for the patterns that are found. We have argued that these macro-level theoretical explanations are the most salient in understanding why it is that temporal and spatial variation in offending occurs. We have also suggested that evolutionary explanations can contribute to our understanding in important ways and hence are relevant for furthering our understanding of the patterns that are found. Indeed, we think that there is substantial scope to build integrated explanatory accounts that incorporate some of the important insights from evolutionary theory—especially those of cultural evolutionary theory—with those developed by mainstream criminologists.

10.7.4 Law-Abidingness Versus Criminality

Crime rates differ widely across different times and different places, for a variety of reasons that are studied by social scientists. But at the level of individuals—that is, if we ask which persons in a given group or a given society are more likely to commit crimes—one clear finding is that there are some significant links between crime and low IQ. On average, persons convicted of criminal offenses score almost two-thirds of a standard deviation unit below average on tests of mental ability (Hirschi & Hindelang, 1977; Wilson & Herrnstein, 1985). But this result raises an interesting question: Is it the case that persons who commit crimes have lower IQs, or is it instead merely the case that persons who get caught for committing crimes have lower IQs?

The evidence suggests that the former is true. One study of about 650 New Zealand teenagers (Moffitt & Silva, 1988) obtained intelligence test scores, anonymous self-reports of delinquent behavior, and police records of arrests and other interventions for delinquency. In this way, the researchers identified three subsets of youths within this sample: A subset of 40 youths who had been in contact with the police because of delinquent behaviors; another subset of nearly 70 youths who had admitted committing equally serious delinquent acts but who had avoided contact with the police; and a final (much larger) subset of youths who had no police contact and no serious self-reported delinquency. Both of the delinquent groups averaged about one-half of a standard deviation lower in intelligence test scores than did the nondelinquent group; there was very little difference between the group that had had police contact and the group that had committed delinquent acts without coming to the attention of the police. Thus, the results of Moffitt and Silva suggest that the lower IQs among persons arrested for crimes are a reflection of a difference between persons who commit and persons who do not commit crimes, not merely of a difference between persons who are caught and persons who are not caught for committing crimes.

Another issue in interpreting the relation between IQ and law-abiding behavior is that of socioeconomic status. Given that lower socioeconomic status tends to be associated, to some extent, with higher rates of crime and with lower IQs, you might wonder whether the negative relation between IQ and crime is attributable simply to low socioeconomic status. In other words, perhaps there is no relation between low IQ and crime if we consider only persons who have grown up in households of equal socioeconomic status.

However, the evidence suggests that this is not the case; instead, persons with higher IQs tend to commit fewer crimes even when socioeconomic status is held constant. One investigation that showed this result was reported by Moffitt, Gabrielli, Mednick, and Schulsinger (1981), who studied a sample of over 4500 young men in Denmark. Of those men, nearly 8% had been convicted of one criminal offense, and another 5% had been convicted of two or more criminal offenses. Using records of these men's socioeconomic status and intelligence test scores as collected years earlier, Moffitt et al. found that the number of criminal offenses was correlated, weakly, with lower socioeconomic status (r = −.11), but somewhat more strongly with lower IQ (r = −.19). When Moffitt et al. controlled for socioeconomic status by calculating a partial correlation (see description earlier in this chapter), number of offenses was still related to lower IQ (r = −.17). This means that even when we consider people who come from households having the same level of socioeconomic status, there is still a modest tendency for those who have lower levels of mental ability to be more likely to commit crimes.

Although the preceding results indicate an association between criminal activity and lower IQs, it is important to keep in mind that this does not mean that persons with high IQs do not commit crime at all. Even though high-IQ individuals are less likely to commit crimes, there is some indication that those who do so tend to “select” crimes that have a higher probability of payoff and a lower probability of arrest. Moreover, there are some “white-collar” crimes—such as large-scale corporate fraud or government corruption—that presumably are committed mainly by high-IQ persons, because only persons with rather high levels of mental ability are able to achieve the positions in which one has the opportunity to commit those crimes.

Why is that persons with higher IQs tend to be more law-abiding? One possibility is that the cost/benefit ratio of criminal activity is higher for persons who have high IQs, because those persons have better chances for educational and occupational success. In contrast, persons with lower IQs may become frustrated by their somewhat poorer prospects for achieving success (Hirschi & Hindelang, 1977). But, more generally, it may also be the case that high-IQ persons are more likely to recognize that committing a crime carries rather high risks of serious penalties and to judge the potential gains from crime as being insufficient to justify those potential losses.

Abstract

Violent crime rates may define perceptions regarding community safety. Data mining and predictive analytics can be used to operationalize Criminal Investigative Analysis, or the behavioral analysis of violent crime. Similar to the use of advanced analytics in other domains, behaviorally segmenting crime based on type, nature, and motive can provide novel, operationally relevant and actionable insight. These methods can be used to support motive determination and enhanced investigative efficacy, as well as crime prevention through information-based approaches to anticipation and influence. Violent crime examples and case studies are provided to illustrate concepts.

9.07.1 Introduction

The crime rate in the United States, particularly among youthful offenders, is growing at an astronomical rate. Of most concern is the increase in the occurrence of youthful violent crime. From 1992 to 1993, the total juvenile arrest rate increased by 5% and juvenile arrests for violent crimes increased 6%. Arrests for weapons violations by youthful offenders rose by 12%, and the arrest rate for homicide increased 14%. At the same time, adult changes in these categories were negligible (Howell, 1995).

Evidence continues to mount that a small proportion of offenders commit most of the serious and violent crimes. It is this small group that captures the public interest in making decisions about policies related to interventions with criminals as a whole. Furthermore, it is known that the onset of serious violent careers begins to increase at age 12, doubles between ages 13 and 14, peaks at ages 16 to 17, drops 50% by age 18, and continues to decrease through age 27. More than half of all violent offenders initiate their violence between ages 14 and 17 and their careers last only one year, with only 4% with a career of five years or more (Elliot, 1994; Elliot, Huizinga, & Morse, 1986). In fact, aggression is a very stable behavioral trait and involvement in violent behavior at an early age is the best predictor of a youth becoming a chronic violent offender. Therefore, to understand violent and chronic offenders, an analysis of this small group of youth must be undertaken.

The purpose of this chapter is to explore the characteristics of children and adolescents who become violent and chronic offenders. It is hoped that by understanding the antecedents to aggressive behavior, a more rational, comprehensive, and exhaustive plan of intervention may be developed. To best discern the attributes of youthful violent and chronic offenders, it is necessary to explore the various systems that contribute to the environments of these juveniles. This will require an analysis of the systems most intimate and internal to the child, namely the biology of chronic, violent offenders, and working outward to the most external spheres of influence on these youth.

International Comparisons

International comparisons of crime rates based on police and court statistics are fraught with difficulties because crime definitions differ, as do police and court practices, statistical conventions, and citizens' reporting of crime to the police.

Victimization surveys using a similar questionnaire in different countries allow a direct way of making offense definitions similar. However, even comparisons of victimization surveys must be done cautiously because of differences in sampling, methodology, and content. Even when questions about victimization are worded similarly, differences in sampling, fielding, and survey methodology, for example, how different known sources of errors are taken into account, may produce great differences.

A series of surveys known as the International Crime Survey (ICS), sometimes also called ICVS, was started in 1989 and has the longest track record as a standardized international victimization survey. There have so far been five main sweeps of the ICS: 1989, 1992, 1996, 2000, and 2004/05. By 2005, these surveys had been done once or more in over 79 different countries (in 38 countries nationwide). They include all member states of the European Union (EU), the United States, Canada, Australia, and Japan as well as many developing countries and transitional countries in Central and Eastern Europe. The ICS gives less detailed and precise information than large-scale national crime surveys. It is also likely that despite efforts to standardize the sampling of respondents and the interview mode, there may be subtle variations between different countries and different sweeps of the survey that may affect comparability. However, one can claim that the ICS is at the moment the best available source of internationally comparative data on general crime.

An illustration of the differences between police statistics and surveys in international comparisons is provided by the ICS, designed to overcome a number of basic comparability problems. The study found that the police recorded crime statistics and the survey rank ordered the jurisdictions more or less identically in the case of car thefts. Auto theft is a relatively well-defined offense that is reported to the police much more frequently that the average offense. With respect to some other crimes, however, there were great differences. For instance, for domestic burglaries the ICS put Norway 60% higher than Finland, while Interpol statistics had 7 times higher rates for Finland than Norway. However, a closer analysis of national police statistics reveals that once incomparable categories are sorted out, this incongruity disappears.

Apart from the previously mentioned ICS, there have been other international surveys on crime victimization. For instance, there have been several international surveys of domestic violence against women. The World Health Organization Multi-Country Study on Women's Health and Domestic Violence Against Women collected data from over 24,000 women from 15 sites in 10 countries representing diverse cultural settings (Bangladesh, Brazil, Ethiopia, Japan, Namibia, Peru, Samoa, Serbia and Montenegro, Thailand, and Tanzania). The prevalence of lifetime victimization of women to physical violence by an intimate partner varied from the high of 61% in the Peruvian province to a low of 13% in the Japanese city. Another study, the International Violence Against Women Survey, IVAWS, was conducted in Australia, China (Hong Kong), Costa Rica, the Czech Republic, Denmark, Greece, Italy, Mozambique, Poland, Philippines, and Switzerland. Both studies have developed the comparative international survey methodology. It is clear that the survey cannot abolish all cultural biases that affect understanding and disclosure of violence, but a good and consistent methodology may ensure that the variations found in prevalence between the sites studied using the same method tend to represent real differences.

According to victimization surveys, satisfaction with police performance correlates highly with the reporting behavior of the population: the more satisfied the population is with the police, the greater proportion of crimes is reported to the police. Satisfaction with the police is higher in the western countries than in the developing countries.

8.1.4 Specific Measure

“Overall, serious crime rate down, but arsons, burglaries, homicides up”3

It is very important to select the specific outcome measure with thoughtful consideration. As the quote suggests, the way that we count crime matters. One of the most popular violence prevention outcome measures is homicide rate. While focus on the homicide rate frequently reflects a significant concern over needless loss of life associated with a violent crime problem in a community, it can be a terrible outcome measure. These numbers tend to be relatively low, which is a good thing for the community, but a challenge from an evaluation standpoint. A homicide also can reflect other factors, including access to timely, competent medical care. Aggravated assaults, on the other hand, are more frequent and often represent incomplete or poorly planned homicides. As such, they represent a good proxy for homicides and are a more effective measure of violent crime.

Similar situations can occur with a variety of measures. For example, arrest-based crime reporting can incorrectly make it look as if crime is increasing in response to a particular initiative. For example, aggressive drug enforcement strategies generally are associated with an increased arrest rate for narcotics offenses. Because arrests are used as the measure of crime, an increased arrest rate can suggest that the problem is getting worse. The truth actually might be that the aggressive enforcement strategy is getting drug dealers off of the street and making the community inhospitable to illegal drug markets, which by almost any standard would be a measure of success. The arrest rate also can be a good process measure, as it definitely shows that folks are out there doing something. Unfortunately, arrest rates can create particular challenges when used as outcome measures. This is not necessarily bad, but it is important to understand what might impact this to ensure that the information is interpreted appropriately and within the proper context.

Deeper understanding of the true goals of a particular intervention also can be used to guide the related performance metrics. For example, there has been an increased deployment and use of closed-circuit television (CCTV) to support security and surveillance efforts in areas deemed to be at high risk for crime and terrorist attacks. Perhaps, the most noteworthy application of this model is the “ring of steel” in London, and more recently the venues for the 2014 Olympics in Sochi, Russia, which include massive deployment of CCTV cameras. The potential value of this model was demonstrated during the London bombings in July 2005. Extensive video footage of the terrorists purchasing supplies, engaging in dry runs, and even planting and detonating the devices associated with the 7/7 attacks in London was associated with brisk investigative pace, including rapid identification and apprehension of the suspects.4 Similarly, surveillance and bystander video footage was used in 2013 to quickly identify the Boston Marathon bombers.5 Unfortunately, they were not able to use the video footage proactively to identify the preoperational planning in support of prevention, thwarting, or consequence management in either of these incidents. So is CCTV in crime prevention a tool, or not? While preoperational planning, to include purchase of the bomb components, rehearsal and dry runs, and even emplacement of the devices was readily apparent in retrospect, the video footage played no role in early identification and prevention or thwarting of the attacks. Therefore, in considering an appropriate metric for CCTV as an effective counterterrorism tool, would it be rapid identification of possible suspects and enhanced investigative efficacy; or prevention, thwarting, consequence management, mitigation, changed outcomes? Metrics matter.

Related to this, scientists at IBM6 posed a similar question regarding the best metric for fraud and financial crimes programs: early detection and a timely response, or anticipation, prevention and influence. In the real world, the best answer probably is both. Ideally, fraud and other patterns of bad behavior will be detected early and prevented or thwarted. In the absence of early intervention, though, the ability to effectively anticipate in support of information-based response and consequence management might be equally important. Again, rigorous evaluation of the outcome will be contingent on selection of the appropriate metric.

Consider how particular measures might change over time. For example, the Project Exile weapon recovery rate rose initially and then fell off as criminals got the message. Similarly, complaint data can change during an intervention. A community experiencing regular drive-by shootings might be somewhat less motivated to report gang-related tagging or graffiti in the area; graffiti might be bothersome, but it pales in comparison to the amount of lead flying through the air each night. As the violent crime rate is addressed, however, and the community becomes reengaged, residents might be more motivated to begin reporting some lesser crimes, which could appear to be an increase in these crimes. Citizens also might be more likely to report random gunfire after an initiative has been established and shown some promise. Prior to deployment of the initiative, there might have been a sense that nothing would change or that there was danger associated with becoming involved. However, as improvements are noticed following the initiative, crime reporting might increase as neighborhoods become revitalized and the residents begin to reengage and participate in the enforcement efforts.

This point is related to a similar issue: All crime is not created equally. How many aggravated assaults equal a homicide? Is an armed robbery equal to a sexual assault? How about a drug-related murder? While these might seem like absurd questions, law enforcement agencies frequently compile and aggregate these numbers and create a composite “violent crime index.” Formerly referred to as “Part I” crimes, these various measures generally are lumped together and used as a generic measure of violent crime in a community. This is unfortunate because combining all this information together increases the likelihood that something important will be obscured.

Many of the crimes frequently included in composite violent crime indices occur with differential frequency. Generally, there are far more aggravated assaults in a community than murders, and far more robberies than sexual assaults. A decrease in a relatively low-frequency crime might be lost when it is considered within the context of all of that additional information. Moreover, these crimes are not equivalent in terms of their impact on a community. Few would argue that homicide is far more serious than an aggravated assault. Why throw them together in a composite violent crime index that weights them equally?

This probably makes sense to most, but an important extension of this issue arises with the use of generic offense categories to evaluate an initiative targeting a specific pattern of offenses. For example, why create a specific model of drug-related homicide if you are going to base the outcome evaluation on the entire murder rate? It is very rare to develop and deploy a crime prevention strategy that addresses everything, even all the crime within a general category. A particular robbery initiative might target street robberies, but the entire robbery rate traditionally is used to evaluate the efficacy of the initiative. Similarly, an initiative targeting commercial robberies is not likely to affect carjackings, but it is very likely that carjackings will be included in the “robbery” outcome measure.

Another point to consider is whether it is even possible to measure what the program is designed to impact. An interesting question emerged out of the Project Exile work: How do we measure the firearms carry rate? This is a very important question, because the stated goal of Project Exile was to reduce the carry rate of illegal firearms. The illegal carry rate, however, would be very difficult, if not impossible, to measure. As a result, additional proxy measures were selected in an effort to measure the efficacy of the program. The proxy measures included the number of illegal firearms recovered, as well as other measures of gun-related violent crime. While it was not possible to accurately measure the true carry rate of illegal firearms, these other measures turned out to be just as important in terms of quantifying community public safety and were linked intrinsically to the original measure of interest.

So, while it might not be possible to directly measure the outcome of interest, there generally are other indicators linked to the original measure that can be documented in its place. For example, investigative efficacy as a measure is likely to be elusive. Case clearance rates, however, can be documented and used as a reflection of an improved investigative process.7 Serious thought to the specific goals of the operational plan and some creativity in the selection of outcome measures can address these challenges, particularly if these decisions are made as part of the operational planning process.

Spatially Referenced Crime Rates

In order to calculate a crime rate, there are two pieces of data that are necessary: crime data and population at risk data, shown in eqn [1]. The crime count is literally the count of criminal events in some defined geographic area. This count is most often based on official crime data, emerging from some aspect of the criminal justice system. The most well-known official crime data are the Uniform Crime Reporting (UCR) data. These data had its beginning in the United States in 1930 and now nearly all law enforcement agencies in the United States provide data to the Federal Bureau of Investigation for the UCR counts (Mosher et al., 2011; FBI, 2012). In Canada, the UCR began later, in 1962, and is gathered by the Canadian Centre for Justice Statistics – responding to the UCR survey in Canada is mandatory. In 1988, a new version of the UCR in Canada was started that added information regarding incidents, victims, and accused persons. This new UCR is referred to as UCR2; equivalent data in the United States are collected under the National Incident-Based Reporting System, that began in 1987 (FBI, 2012; Statistics Canada, 2012). Another form of official crime data is calls for service data from the police. Though often considered unofficial because these data are not dependent upon a criminal charge and often represent police activity, these are very useful data because a geographic location and time is most often included and the calls for service can be separated to only include the calls that the police identify as a criminal event (Sherman et al., 1989).

[1](Crime countPopulation at risk )×Scalar

Despite the widespread availability of these official crime data, there are a number of limitations. First, crime reporting can vary from police detachment to police detachment. This may be due to factors such as the local policing and/or population culture considering factors such as (a lack of) tolerance on particular issues. Second, not all criminal events are reported to the police. In Canada, only 31% of criminal victimization was reported to the police in 2009, down from 34% in 2004 and 37% in 1999 (Perreault and Brennan, 2010). This is of critical importance in spatial criminology because if the spatial patterns of crime for official data are different from the spatial patterns of crime more generally, all inference from official data may be spurious. However, very little can be done regarding these first two limitations.

A third limitation is in regard to the need for population at risk data to calculate the crime rates. In spatial criminology, that often uses the census tract as its spatial unit of analysis, the most common source of population at risk data is the census – resident population. These data are very easy to obtain and high quality, but there are two issues with using these data. First, censuses are only undertaken every 5 (Canada) or 10 years (United States and United Kingdom) in most countries. Consequently, depending on the date of the crime data obtained there may be as many as 9 years discrepancy between crime count and population at risk data. If populations are stable over long periods of time this is not problematic, per se, but in cities that grow this could lead to significant overestimates of crime rates if crime counts are for a later year than the resident population.

The second issue with using the resident population from a census as the population is risk is whether or not it is an accurate representation of where people are. This is a critical question in spatial criminology because if the crime rate is to represent an overall measure of risk (Boggs, 1965) the variable representing the population at risk should measure where people actually are. Recent research on this issue has shown that the ambient population can be quite different from the resident population leading to changes in inference regarding spatially referenced crime rates (Andresen, 2006, 2013; Andresen and Jenion, 2010). In this work, data from the Oak Ridge National Laboratory was used to measure the ambient population that was defined as the number of people in a square kilometer averaged over a 24-h period for any typical day of the year – see Andresen (2006) for details of these data.

In an investigation of the similarity between maps of resident-based and ambient-based crime rates, Andresen and Jenion (2010) found that, despite the visual appearance of similar spatial patterns and a statistically significant relationship, resident-based crime rates were very poor predictors of ambient-based crime rates. Though we do not discuss this prediction, this is what we assume when we use the resident population: it may not be the best measure, but it is representative of the true population at risk. Moreover, places that attract populations throughout the day, measured using the ambient–resident population ratio, are commercial/shopping areas and major transportation routes.

In subsequent research, Andresen (2011, 2013) showed that ambient-based crime rates exhibited significant differences to resident-based crime rates in the context of local analysis, local Moran's I. In this work, not only did the spatial pattern of clustering change with some units of analysis changing statistical significance, but also when statistically significant in both analyses the classifications in some places would be completely different – most often, high crime areas remained high crime areas and low crime areas remained low crime areas, but the units of analysis contained within each of those classifications changed. In a spatial regression context, the similarity of results for resident- and ambient-based crime rates differed on the year of data and the size of the unit of analysis. In some cases, the results from the ambient-based crime rates outperformed the resident-based crime rate results in the context of: greater (pseudo) R2 values, more reasonable parameter estimates, more statistically significant relationships, and a better concordance with theoretical expectations. In other cases, there were very few differences with limited consequences to the results.

Overall, this research shows that there are good theoretical reasons to question the use of the commonly used resident population in spatially referenced crime rates – crime rate calculations for what may be considered a ‘closed system’ of population movement such as a metropolitan area or a subnational unit (province or state) will most probably be unaffected. Though there may not necessarily be an impact on the results, such an impact is definitely possible and may alter any interpretations. Moreover, just because there is no impact for 1 year of data and 1 spatial unit of analysis, an impact may be present for a different year of data and/or a different spatial unit of analysis even in the same city.

Aggregate Police Presence and Crime

Studies of police hiring and crime rates have been plagued by a number of impediments to causal inference. Among these are cross-jurisdictional differences in the recording of crime, feedback effects from crime rates to police hiring, the confounding of deterrence with incapacitation, and aggregation of police manpower effects across heterogeneous units, among others. Yet the challenge that has received the most attention in empirical applications is the simultaneity problem referred to in the previous section – in the present case, the feedback from crime rates to police hiring.

The two studies of police manpower by Marvell and Moody (1996) and Levitt (1997) are notable for their different identification strategies. The Marvell and Moody (1996) study is based on an analysis of two panel data sets, one composed of 49 states for the years 1968–93 and the other of 56 large cities for the years 1971–92. In order to overcome the problem of a bidirectional relationship between police manpower and the crime rate, they examine the effect of police manpower in the prior year on the crime rate in the current year. The strongest evidence for an impact of police hiring on total crime rates comes from the city-level analysis, with an estimated elasticity of −0.3, meaning that 10% growth in police manpower produces a 3% decline in the crime rate the following year, representing a modestly large effect.

However, regression analyses of this type do not generally provide a valid basis for making causal claims. But other forms of analysis can provide such a basis. One is an approach that exploits pseudorandom variation in policies that are only correlated with the crime rate through their effects on police manpower, in what is known as an instrumental variables (IVs) regression. Levitt (1997) performs this type of analysis from a panel of 59 large cities for the years 1970–92. Reasoning that political incumbents have incentives to devote resources to increasing the size of the police force in anticipation of upcoming elections, he uses election cycles to help untangle the cause–effect relationship between crime rates and police manpower. Levitt's model produces a very large elasticity of about −1.0 for the violent crime rate and a more modest but still meaningful elasticity of −0.3 for the property crime rate. Following Levitt's use of the electoral cycle as an instrument for the number of sworn police officers, other studies have employed alternative IVs and reported comparable elasticities.

In recent years, a number of more targeted tests of the police–crime relationship have appeared. These studies investigate the impact on the crime rate of reductions in police presence and productivity as a result of massive budget cuts or lawsuits following racial profiling scandals. Each of these studies concludes that increases (decreases) in police presence and activity substantially decrease (increase) crime. By the way of example, Shi (2009) studies the fallout from an incident in Cincinnati in which a white police officer shot and killed an unarmed African-American suspect. The incident was followed by 3 days of rioting, heavy media attention, the filing of a class action lawsuit, a federal civil rights investigation, and the indictment of the officer in question. These events created an unofficial incentive for officers from the Cincinnati Police Department to curtail their use of arrest for misdemeanor crimes, especially in communities with higher proportional representation of African-Americans out of concern for allegations of racial profiling. Shi demonstrates measurable declines in police productivity in the aftermath of the riot and also documents a substantial increase in criminal activity. The estimated elasticities of crime to policing based on her approach were −0.5 for violent crime and −0.3 for property crime, both of which have policy significance.

The ongoing threat of terrorism has also provided a number of unique opportunities to study the impact of police resource allocation in cities around the world. The study by Klick and Tabarrok (2005) examines the effect on crime of the color-coded alert system devised by the US Department of Homeland Security (in the aftermath of the 11 September 2001, terrorist attack) to denote the terrorism threat level. Its purpose was to signal federal, state, and local law enforcement agencies to occasions when it might be prudent to divert resources to sensitive locations. Klick and Tabarrok (2005) use daily police reports of crime for the period of March 2002–July 2003, during which time the terrorism alert level rose from ‘elevated’ (yellow) to ‘high’ (orange) and back down to ‘elevated’ on four occasions. During high alerts, anecdotal evidence suggested that police presence increased by 50%. Their estimate of the elasticity of total crime to changes in police presence as the alert level rose and fell was −0.3.

To summarize, aggregate studies of police presence conducted since the mid-1990s consistently find that putting more police officers on the street – either by hiring new officers or by allocating existing officers in ways that put them on the street in larger numbers or for longer periods of time – has a substantial deterrent effect on serious crime. There is also consistency with respect to the size of the effect. Most estimates reveal that a 10%increase in police presence yields a reduction in total crime in the neighborhood of 3%, although studies that consider violent crime tend to find reductions ranging from 5 to 10%. Yet these police manpower studies speak only to the number and allocation of police officers and not to what police officers actually do on the street beyond making arrests. The next section proceeds from here by reviewing recent evaluations of deployment strategies used by police departments in order to control crime.

Lower crime rates

The effect of education in reducing crime rates and criminal justice system costs has received more attention than other externalities. Witte's (1997) review reveals that further education of those who have started on a life of crime is of limited effectiveness, whereas reducing high-school dropout rates and increasing 2-year college enrolments that cause young males to be under supervision in school (and in employment later) are effective. The value of high-school or college graduation in reducing murder rates (violent crime) and property crime (all other crime) after controlling for per capita income, lagged unemployment, inequality, and poverty is estimated to be $719 per year per graduate for lower murder rates, and $4928 per year per graduate for all of the many other kinds of crime. Higher education contributes to white-collar crime, a negative externality, but this has been netted out against education's positive benefits in reducing overall crime. Lochner and Moretti (2002) do not control for income.

5.3 Neighborhood context: Othering and belonging

While feelings of insecurity arise from crime rates and the danger of the unknown, barriers to travel also exist for people who draw suspicion because they are judged not to belong in a particular place. One explanation for the pervasive nature of racial profiling in police stops is the racial threat hypothesis. The hypothesis suggests that elites seek to impose greater control on the Black population as it grows because of the political, economic, and criminal threats they pose (e.g., Eitle et al., 2002). In some places, there is evidence that this theory may hold for traffic stops, where more stops occur in areas with more Black or Latino residents (Ingram, 2007; Roh and Robinson, 2009). But others have found compelling evidence for a different “race-out-of-place” theory, in which drivers who are assumed to be interlopers to the neighborhood are treated as more suspect. Scholars have found that Black drivers are more likely to be stopped in predominately white neighborhoods and vice versa for white drivers (Novak and Chamlin, 2012; Rojek et al., 2012). An version of this theory grounded in economics is the postindustrial policing thesis, which suggests that gentrifying neighborhoods or cities with growing creative class economies will increase social order policing for relatively minor infractions of the law (Laniyonu, 2018; Newberry, 2021; Sharp, 2014).

The implications of not belonging in a neighborhood are increased surveillance of people of color by both police and private citizens on the one hand, and their suspicion of planning efforts on the other, heightened by the twin concerns of gentrification and displacement. Particularly with respect to cycling and recreational facilities, communities of color have viewed new infrastructure with skepticism because they perceive the investment to have come only after economic redevelopment of their neighborhoods had begun (Hoffmann and Lugo, 2014; Lubitow and Miller, 2013). Mainstream bicycle advocacy, too, has promoted cycling from the perspective of the young, middle-class, white professional returning to the city, while ignoring the circumstances of people of color and laborers who have relied on cycling in the absence of safe infrastructure (Hoffmann, 2016; Lugo, 2018; Sheller, 2015). These histories have manifested in differences in the use of such facilities. Marginalized cyclists, for example, have substituted “human infrastructure,” or their accumulated knowledge of and experiences with the city, to navigate when hard infrastructure was non-existent, relying on paths they know to be safe even if designated routes were built (Lugo, 2013). In the case of new recreational facilities, people of color who cycle or walk along paths self-segregate away from gentrifying neighborhoods to avoid inviting suspicion and run-ins with the police (Harris et al., 2020a,b).

The perception of non-auto users as “others” is also evident in public transit in the United States. There are both income and race divides both between public transit riders and users of other modes and within public transit modes themselves. Data from the 2017 National Household Travel Survey show that transit riders are on the whole poorer than users of other modes, while bus riders have a median household income below poverty wages compared to six-figure household incomes for rail riders. The majority of transit riders are people of color, compared to less than half of all other modes. Again, there are stark differences between bus riders and rail riders: people of color make up almost three-quarters of bus riders, compared to almost one-half of subway and light-rail riders and one-third of commuter rail riders (Federal Highway Administration, 2017). These demographic patterns are hard to disentangle from a general stigma against public transportation (Schweitzer, 2014); some neighborhood residents oppose new transit service because they fear the “other” who rides it (Weitz, 2008), while transit riders of color themselves perceive discrimination on the part of other riders or transit officials as barriers to use (Barajas et al., 2018; Liu and Schachter, 2007; Lubitow et al., 2017).