View this issue as a PDF: CrimHighlightsV17N2.pdf
This issue of Criminological Highlights addresses the following questions:
- Are complex algorithms for predicting recidivism helpful?
- Is the presence of substance abuse treatment programs in a community associated with changes in crime rates?
- How does contact with the criminal justice system affect the mental health of an accused person?
- What do the police need to do to encourage Muslim communities to report suspected terrorism activities?
- What can we learn from the failure to replicate an experiment involving the police?
- What can be done to help people with criminal convictions find housing?
- How does pretrial detention affect crime?
- What should we do to reduce police shootings of civilians?
A sophisticated looking, commercially marketed risk prediction instrument is no more accurate in predicting criminal reoffending than ordinary people’s risk predictions using age, sex, and readily available indicators of a person’s criminal history.
Complex algorithms, using many data points, to predict recidivism (or offending while on pretrial release) are attractive to criminal justice professionals because they give an apparently objective answer to the question, “Is it safe to release this person into the community?”
This study provides a detailed analysis of one such prediction system, COMPAS, which was developed by Equivant (previously Northpointe). The details of the algorithm used by COMPAS are not public.
There is little question that such systems are “valid,” in that they predict better than chance. The more important questions are what kinds of errors they make and whether they are more accurate than less sophisticated systems.
This paper uses data from 7214 defendants in one county in Florida and compares the accuracy of the predictions made by COMPAS, using an undisclosed number of features, to two other kinds of predictions: (a) Ordinary statistical predictive models using a small number of characteristics of the person, and (b) Intuitive predictions made by ordinary people who have no particular expertise in predicting recidivism. There are disputes in the research literature concerning the best measures to use to describe the accuracy of a predictive scale. This paper presents a number of them that, fortunately, are readily interpretable.
The results are remarkably consistent. Overall accuracy for COMPAS predictions of recidivism (arrest within two years) was indeed better than chance. But its accuracy was no different from simple logistic regression models using age, sex, charge, and some features of the criminal record. Perhaps most interesting is that ordinary people who volunteered to do a web-based survey and were presented with summaries of information about these Florida defendants that included only sex, age, offence, and 3 features of their criminal record, performed just as well as COMPAS.
Though COMPAS was more accurate than chance (65.2% accuracy overall), and was equally accurate for black and white defendants, the types of errors made by the software were different for the two racial groups. False positives (predicting recidivism when it didn’t occur) were much more common for black defendants than for white defendants. On the other hand, false negatives (predicting recidivism would not occur when it did) were more common for white defendants. These same differences were found for untrained people in the web-based survey and for an additional sample of ordinary people who were given the defendant’s race when making their predictions. In other words, the commercial software disadvantaged black defendants – and gave advantages to white defendants – in a manner that was very similar to the manner in which ordinary people’s risk assessments disadvantaged black defendants.
Conclusion: “When considering using software such as COMPAS in making decisions that will significantly affect the lives and well-being of criminal defendants, it is valuable to ask whether we would put these decisions in the hands of random people who respond to an online survey because, in the end, the results from these two approaches appear to be indistinguishable” (page 3).
Reference: Dressel, Julia and Hany Farid (2018) The Accuracy, Fairness, and Limits of Predicting Recidivism. Science Advances (American Association for the Advancement of Science), 4.
The presence of substance abuse treatment facilities in a community reduces crime in that community.
The purpose of providing treatment for those addicted to various substances is not primarily to reduce crime. However, community programs that address social and personal problems for those with substance abuse problems might well be expected to reduce crime simply because they improve the well-being of those in the community (Criminological Highlights 17(1)#1).
Drug addiction problems do not happen in isolation. One of the possible benefits of having substance abuse treatment (SAT) facilities readily available in a community is that these organizations can refer their clients to other programs (e.g., for mental health treatment) that clients might not otherwise connect with. This paper examines the influence of SAT facilities at the (US) county level, not on individuals, but instead on a set of indicators of the well-being of the community: crime rates. Hence the study “cannot separate the effects of SAT facilities on those who receive treatment from the effects of SAT facilities on the broader community” (p. 4).
From a national data set in the US, it was possible to estimate the number of SAT clinics in a given county. The study examined counties in which there was at least one SAT in the period 1999-2010 and looked at the relationship of changes in the concentration of SATs to changes in homicides, violent crimes, and financially-motivated crimes. Various controls (e.g., unemployment rate, police officers per capita, age and racial makeup of the county) were included in the analyses. The primary analyses examined the effects of the number of SATs in the community in one year on crime in the next year.
Not surprisingly, an increase in the concentration of SATs was associated with a net reduction in drug-induced mortality rates. As the concentration of SATs increased, there tended to be a reduction in homicides, though there was no apparent effect of SATs on intra-familial homicides. More generally, there was a reduction in various forms of violent crime associated with an increase in the concentration of SATs. The exception to this was simple/common assaults (which constituted about 77% of all violent crime): There was no association between an increased presence of SATs and simple/common assaults. However, an increased concentration of SATs was associated with a small reduction in financially-motivated crimes (when excluding the most minor forms).
Given the design of the study, there is always the possibility that the causal relationship is in the opposite direction: Facilities may move into locations where crime is coming under control. To test for this, an analysis was carried out in which crime rates were used to predict the number of SAT facilities in a future year. These effects were never significant, suggesting that it is the increased presence of SATs that may be responsible for a lower crime rate.
Conclusion: The effects of placing additional substance abuse treatment facilities in a county were consistent. With the exception of minor crimes (minor assaults or simple thefts), additional treatment facilities in the community in a given year tended to reduce more serious crime in the following years. However, though the effects were statistically significant, they were not large in size. Hence it is unlikely that the crime reduction impact of these substance abuse facilities would be noticed in a community. Nevertheless, given that the purpose of these facilities is to address addiction issues (and deal with associated problems), the crime reducing effect is a favourable unforeseen outcome of providing a much needed service to people in need.
Reference: Bondurant, Samuel R., Jason M. Lindo, and Isaac D. Swensen (2016). Substance Abuse Treatement Centers and Local Crime. National Bureau of Economic Research (Cambridge, MA, USA) Working Paper 22610.
Being arrested and being incarcerated each have negative impacts on the mental health of young adults.
Arrest and incarceration repeatedly have been shown to have negative impacts on a person caught up in the criminal justice system and on others related to that person (e.g., Criminological Highlights 17(1)#6),16(3)#2, 16(2)#1, 16(4)#5,#7, 16(6)#8) 14(2)#1). This paper examines the impact of criminal justice contact on the self-rated mental health over a ten year period of a sample of people age 18-20 at the start of the study.
Arrest could have a negative impact on people’s mental health in a number of ways. In addition to stigmatizing effects, arrest may lead to feelings of powerlessness and alienation. Arrest also may disrupt normal obligations (e.g., work or child care) and lead to uncertainty about the future. Conviction and incarceration can have effects above and beyond the effects of arrest.
Obviously, those arrested (and perhaps subsequently incarcerated) are different from those who are not. This paper addresses this problem by looking at changes over time in self-reported mental health in a sample of about 7500 people who were interviewed every second year. In that way, it was possible to see if there were negative impacts on the mental health of a young person after the experience of arrest (and incarceration, if that occurred).
Mental health was measured with 5 questions about how often, in the previous month, respondents felt nervous, calm and peaceful (reverse coded), downhearted and blue, happy (reverse coded) and ‘so down in the dumps that nothing could cheer them up’ (p. 725).
Arrest was associated with a substantial subsequent decrease in self-rated mental health, even when controlling for 13 factors that change over time (e.g., age, marital and family status, employment, economic situation, drug or alcohol abuse and delinquent behaviour). Incarceration also was related to a decrease in self-rated mental health, above and beyond the impact of arrest (including these same control variables). The simple effect of incarceration (above and beyond the effect of arrest) was, however, smaller than the effect of arrest.
Keeping in mind that this sample was relatively young and followed for about 10 years, it was possible to examine the impact of the first incarceration. When one looks at the impact of the first arrest and first incarceration, the negative impacts of these two criminal justice interventions were comparable in size. A more detailed analysis, however, suggested that it was pretrial detention (incarceration) rather than sentenced incarceration that contributed to the decline in self-rated mental health.
To strengthen the conclusion that the effects were due to criminal justice intervention causing a decline in mental health (rather than mental health problems leading to arrest and incarceration), a separate analysis determined that there was no significant impact of changes in mental health status on subsequent arrest and incarceration.
Conclusion: These data suggest that arrest and pre-trial detention – but not conviction – have independent negative impacts on the self-rated mental health status of young adults. Arrest, however, accounts for over half of the negative association between incarceration and mental health. These negative effects are fairly similar across racial/ethnic minority groups.
Reference: Sugie, Naomi F. and Kristin Turney (2017). Beyond Incarceration: Criminal Justice Contact and Mental Health. American Sociological Review, 82(4), 719-743.
An Australian study demonstrates that the police can affect the likelihood that Australian Muslims will report potential extremist activities to the police by treating them in a manner that engenders trust in the police.
For many years, and most clearly since 2001, western countries have had concerns about whether members of certain groups – most notably Muslims – are willing to cooperate with police in criminal investigations. If the police are going to be effective in preventing possible terrorist acts (or other crimes), they need to know what is happening in the communities they police and they need to be able to rely on people’s willingness to report potential terrorist activities to police. This study examines “how willing [Australian] Muslims are to work collaboratively with police in the fight against terrorism” (p. 1145).
This study builds on previous work (e.g., Criminological Highlights 11(4)#1) in examining the role of procedurally just policing on Muslims’ willingness to report potential extremist activities to the police. It extends that work by examining possible intervening mechanisms: the perception of the legitimacy of the local police and the extent to which respondents identify themselves as members of their national community.
A complex identification and screening process was used to locate 800 Australian Muslims in three cities. They were interviewed in person in 2014 and were asked 10 questions related to procedural justice in dealing with issues surrounding terrorism (e.g., “When dealing with people in your community concerning issues of terrorism, the police…give people a chance to express their views before making decisions”, “…treat people with dignity and respect”, etc.). To assess the perceived legitimacy of the police they were asked 8 questions (e.g., agreement with “The way the police act in my community is usually consistent with the values of people in my community about what is right and wrong”). To identify how much they identified as Australians, they were asked 4 questions (e.g., “I see myself as Australian first and Muslim second”). Finally people were asked how likely they were to report to the police 6 different activities (e.g., “A person talking about travelling overseas to fight with Muslims”; “A person reading religious literature you believe to be extremist”). Control variables included age, gender, education, income and the frequency of contact with the police in the previous 12 months.
In a regression model, the strongest single predictor of self-reported cooperation with the police was the respondent’s view of how procedurally just the police were. Above and beyond this, perceptions of police legitimacy and identification as an Australian also contributed to an understanding of who would report suspicious activities to the police, though each of these effects is weaker than procedural justice. Those who had contact with the police in the previous 12 months were less likely to cooperate with them.
But in addition, “[i]f police are seen to be using procedural justice when engaging Muslims, they are viewed as more legitimate and [the respondent’s] social identification with Australia is bolstered” (p. 1156). However, a somewhat different analysis suggested that perceived procedural justice had its effect on cooperation with the police because it increased the perceived legitimacy of the police, rather than by increasing a respondent’s identification as an Australian.
Conclusion: The police often rely on members of the public for key pieces of information about suspicious activities in their community. Potential terrorism activities are one such example, but not the only one. This study adds to a growing list of research (e.g., Criminological Highlights 4(4)#1, 7(1)#4, 11(4)1, 12(5)#2, 15(1)#3, 15(3)#2, 15(4)#3) suggesting that procedurally just behaviour on the part of police can be important in ensuring cooperation and compliance with the police. This study demonstrates that the police can increase their legitimacy and their effectiveness in combatting terrorism simply by acting in a fair and just manner with those in the community who might have very useful information.
Reference: Madon, Natasha S., Kristina Murphy and Adrian Cherney (2017). Promoting Community Collaboration in Counterterrorism: Do Social identities and Perceptions of Legitimacy Mediate Reactions to Procedural Justice Policing? British Journal of Criminology, 57, 1144-1164.
An attempt at replicating a procedural justice experiment with Scottish police engaged in a road safety campaign did not lead to the expected results. An investigation of the program’s failure suggests that insufficient attention was paid to implementing it properly.
It is not unusual that something that appears to ‘work’ in one setting does not work in another. There are several possible explanations for these “failures to replicate.” This study investigates an attempt, in Scotland, to replicate an Australian study in which police tested the impact of a set of “procedural justice” approaches in their interactions with citizens during ordinary police-initiated encounters.
The original Australian study found that members of the public who interacted with police officers during an ordinary police-initiated encounter were more favourable to the police when the police spoke to the citizen in a manner that “signaled dignity and respect, invited citizen participation, and provided clear explanations during encounters [with the police]” (p. 323).
In the Scottish replication, half of the Scottish road police units were randomly assigned to implement this new model of interaction (the ‘experimental group’). This interaction was supposed to include “all of the key elements of the procedural justice model” (p. 323). The other half of police units (the ‘control group’) were to conduct business-as-usual. Drivers were, at the end of their interaction with the police (in both the experimental and control group), given a questionnaire to fill out giving their assessment of the interaction. When completed, the questionnaire was to be returned by mail.
Implementation was not smooth. Though the original intent was to involve all officers in the experimental group in a briefing section, this turned out to be impossible. Instead, all officers in the experimental group were given written instructions on the key messages they were to communicate to the citizen during the interaction. These instructions included an actual example, but “the intention was not that the officers would follow a given script verbatim” (p. 324). The results were exactly the opposite from what was expected: in the ‘control’ areas, judgements of the fairness of the police and satisfaction with them improved in comparison to pre-intervention data; in the experimental areas, ratings of the police did not improve.
There are a number of plausible reasons for the failure. Interviews with 55 officers in the experimental group revealed a number of implementation problems. It appeared that there was “a reliance on top-down command to motivate and regulate officer involvement” (p. 326). Some officers felt they had to follow the instructions verbatim rather than incorporate aspects of procedural justice into their interaction with the public. Other officers had no idea they were supposed to do anything special. Still others were aware of the directive, but felt they were already doing what the directive told them to do; hence they could ignore the directive. More generally, the officers didn’t seem to believe in the purpose of the experimental change. All of this “led to a very negative set of reactions across the experimental units and the different ranks of officers within the units” (p. 334). This is not surprising since “Many officers felt that the experimental requirements, as they had understood them, would have a damaging effect on how they were perceived by the drivers [they were interacting with]” (p. 335). Others saw it as part of “long term erosions to officers’ discretion” (p. 336).
Conclusion: This experimental program on procedural justice was implemented without adequate external monitoring and without officer buy-in of the essential ideas. Hence its implementation may have “triggered a negative response among the officers” (p. 326). Clearly normal “internal structures and processes cannot be relied on to disseminate and communicate details [of programs like this]” (p. 341) especially when their purpose is to get police to behave differently form the way they normally do. In addition, implementation needs to take into account the particular culture and practices of the organization (the police in this case) that may be suspicious of ideas that originate from outside of the organization (researchers in this case). Third, it would be helpful if there was strong support for the assumptions behind the program. Clearly new ideas in policing need to be tested. But the details of how to carry out such tests need to be thought out as thoroughly as the ideas themselves.
Reference: MacQueen, Sarah and Ben Bradford (2017). Where Did It All Go Wrong? Implementation Failure – and More – in a Field Experiment of Procedural Justice Policing. Journal of Experimental Criminology, 13, 321-345.
The effect of a conviction for marijuana possession on obtaining rental housing depends on how old the conviction is, whether the government says it should be ignored in employment situations, and whether the conviction was for a felony or a misdemeanor.
Having a criminal record has been shown to decrease the likelihood that a person will find employment (Criminological Highlights 6(3)#2, 15(1)#7). But having a record almost certainly interferes with ordinary life in other ways. This paper looks at the manner in which different kinds of criminal records – apparently received for similar offences – affect an important aspect of integrating into society: obtaining rental housing.
Some earlier research has demonstrated that having a criminal record reduces the likelihood that a landlord will consider renting to prospective tenants. This paper compares the effects of different kinds of records. The experiment described in this paper was carried out by making telephone calls under the name “Matthew O’Brien” to rental property managers who had advertised housing for US$1200/month or less (modal rent: $700) in 20 neighbourhoods in Columbus, Ohio. Previous studies have shown that almost everyone without a criminal record is considered for tenancy, Therefore, because this study was interested in the effects of different types of records, the ‘applicant’ always identified himself as having a criminal record.
The ‘baseline’ criminal record was described by the ‘applicant’ as follows: “I have a felony drug conviction for marijuana possession that is one year old… I don’t have any other convictions or anything else pending.” 64% of the property managers turned the applicant down on the basis of that information. Each of the other descriptions of criminal record led to increased success in being accepted as a possible tenant. If the criminal record was described as a misdemeanor, only 16% turned the applicant down. If the record was a ‘ten year old felony’ 44% turned the applicant down. As time since the conviction increased, the impact of the record on obtaining housing declined somewhat. But it is interesting to note that a ten year old felony conviction (with no subsequent convictions) is, from the perspective of getting housing, more of an impediment than a 1-year old misdemeanor conviction for the same basic offence.
Perhaps the most interesting finding is that if the applicant with a one year old felony conviction reported that he had a “Certificate of Qualification for Employment” (CQE) from the state government, only 32% turned him down (compared to 64% without a CQE). A CQE “was designed to demonstrate rehabilitation to employers, remove automatic licensing bars, and protect employers from negligent hiring claims” (p. 528). The findings in this study show that being certified by the state to be employable made those with convictions more eligible for housing. However, the applicant with a 1-year old felony and the CQE was still more likely to be turned down than the applicant with a 1-year old misdemeanor.
Conclusion: Having a recent conviction and a felony (rather than a misdemeanor) criminal record makes it more difficult to obtain one of the fundamental requirements for successful reintegration into society – housing. However, even though criminal records may be public in some jurisdictions and even though it may not be possible to ban landlords and property managers from asking potential tenants about criminal records, it seems that formal state processes that deem the person to be no longer deserving of special (negative) treatment because of their past behaviour can have a significant effect on lessening the effect of this stigma.
Reference: Leasure, Peter and Tara Martin (2017). Criminal Records and Housing: An Experimental Study. Journal of Experimental Criminology, 13, 527-535.
Pretrial detention of ordinary offenders appears to have some long-term negative impacts on those detained while not providing any long term public safety.
Many countries, including the US and Canada, have large numbers of people who are detained prior to trial. In the US, it is estimated that about 23% of those imprisoned on an average night are awaiting trial. In Canada, the figure is 37%. This paper examines what the impact of detention is on those being detained.
It is safe to assume that those held in pretrial detention are different from those who are released. Similarly, it is safe to assume that the charges that detainees face, and the evidence against them, also differ from those who are released. The challenge, then, is to find comparable groups of people who are either released or held prior to trial. This paper exploits a well-established criminal justice fact: judicial officers vary in the leniency of their decisions. Furthermore, in the two counties in which this study was carried out, cases were effectively randomly assigned to judges. Judges at the bail stage in these jurisdictions have limited information on which to decide on release or detention. For cases in the middle of a ‘seriousness’ dimension – where it is plausible for the accused either to be detained or release – the essentially random assignment of cases to judges means, effectively, that accused people are randomly assigned to have either a high probability of release or a high probability of being held. There were substantial differences in the tendency to release accused people across judges in each jurisdiction. This study focuses on those cases in the middle where judges disagree on whether an accused should be released or detained. The judges doing bail were different from those making other decisions about the accused.
Those who were lucky enough to get a lenient bail judge (i.e., they were released pre-trial) were less likely to be found guilty (and less likely to plead guilty). Not surprisingly, in comparison to those not released (or released after being held for more than 3 days) those released within three days were more likely to fail to appear in court before trial. More importantly, however, there was no overall impact of being detained or not on the likelihood of committing an offence within 2 years of the bail hearing. This pattern reflects two separate findings. First, those released prior to trial were, not surprisingly, more likely than those detained to commit a new crime prior to disposition. Second, however, those released were less likely to commit a new crime after the original case was disposed compared to those who were detained. Overall, then, the community receives short term benefits of pre-trial detention. But the short-term benefit is wiped out by the fact that those detained are more likely to offend within 2 years of the bail hearing.
The researchers were able to link these defendants to federal tax returns. The tax data showed that defendants who were initially released at their bail hearings had higher formal employment and earnings 3-4 years after the bail hearing than those who were detained. This effect probably comes about in part because those not subjected to pretrial detention are less likely to be found guilty and, as a result, are able to obtain employment. This may also subsequently reduce the likelihood of offending.
Reference: The results of this study – in which criminal defendants on the basis of a near-random decision were detained or released at their bail hearing – suggest that “the adverse labour market outcomes and criminogenic effects [of being processed in the criminal justice system] begin at the pretrial stage prior to any finding of guilt, highlighting the long-term costs of weakening a defendant’s negotiating position before trial and the importance of bail in the criminal justice process” (p. 236).
Reference: Dobbie, Will, Jacob Goldin, and Crystal S. Yang (2018). The Effects of Pretrial Detention on Conviction, Future Crime, and Employment: Evidence from Randomly Assigned Judges. American Economic Review, 108(2), 201-240.
If we were truly interested in reducing the number of fatal police shootings of civilians, we might shift our attention from deterrence models that search for blame to prevention models that search for rare combinations of risk factors and errors that produce fatal shootings.
When police killings of citizens take place and the police officer who fired the gun is not charged with or not found guilty of a crime, the incident is often characterized as “a tragic failure to hold an individual police officer accountable” rather than even considering the possibility that, in addition, the death was “a failure to hold organizational systems accountable” (p. 422). Said differently, it is important that police be held accountable for their actions, but if we wish to reduce police killings, we should broaden our approach to understanding and intervening to reduce the size of the problem.
There are reasons to believe that the number of people killed as a result of officer-involved shootings (currently about 1000 per year in the US) will not be reduced solely by punishing police. These reasons include the fact that convictions are hard to obtain, the high esteem in which police are held by the public, and the fact that many killings do not involve criminal intent or the breaking of police rules. Police shootings in the US occur primarily in smaller cities: 69% of police shootings in the US took place in cities of 250,000 or less. About half occurred in cities with 50,000 or less.
The number of police shootings has been reduced before. Between 1970 and 1985, the number of civilians killed by police was cut in half, in large part because of bans on shooting nonviolent fleeing suspects. Restrictions on when police could shoot were shown (in the early 1980s) to have “no adverse effects on police officer safety and crime, but… large reductions in police shootings did follow [the implementation of these restrictions]” (p. 426). These restrictions, to be effective, had to be supported internally, implemented appropriately, and enforced.
After 1985, the number of police killings rose in part, perhaps, because police were increasingly armed with semiautomatic firearms with large ammunition clips. More recently, attempts to reduce police shootings of civilians (e.g., through mandatory training in de-escalation tactics) have never been evaluated properly. But a fair amount is known about police shootings (see Zimring’s 2017 book When Police Kill). What is needed to fully to use this information effectively is what has been called “preventive imagination” (p. 441). Zimring notes that if one wants to save lives, four approaches can be taken: look for ways to reduce shooting incidents, reduce the number of shots fired, provide immediate medical attention, and transport the victim immediately to a trauma centre.
A broader approach to police killings of civilians would be to examine how other (safety) systems learn from rare catastrophic events (such as plane crashes). Shifting attention from a ‘blame culture’ to a ‘learning culture’ would focus on what organizations can do to reduce harm. “It is the complexity of the high-risk systems that causes extreme harm…. Fatal police shootings shine the spotlight on the shooter rather than the complex organizational processes that recruited, hired, trained, supervised, disciplined, assigned, and dispatched the shooter before anyone faced a split-second decision to shoot” (p. 434). Focusing on ‘who is to blame’ distracts attention from the manner in which a complex interacting system might have failed.
Conclusion: This paper does not offer simple solutions to police shootings, but it does suggest there is need to move beyond the view that simple deterrence (through punishing shooters) will solve the problem. And it suggests there is value in looking at shootings as failures of police systems rather than just failures of individuals. Fixing systems – involving everything from recruitment and training to the dispatching of officers – may, in the long run, provide more adequate solutions. “Policing so that all lives actually matter… requires that police, scholars, and the public go back to the drawing board to design a system specifically aimed at placing preservation of life on an equal (and often higher) level with swift enforcement of the law” (p. 441).
Reference: Sherman, Lawrence W. (2018) Reducing Fatal Police Shootings as System Crashes: Research, Theory, and Practice. Annual Review of Criminology, 1, 421-449.