Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives I examine housing instability among individuals with a felony conviction but no incarceration history relative to formerly incarcerated individuals as a means of separating the effect of felon status from that of incarceration per se—a distinction often neglected in prior research. I consider mechanisms and whether this relationship varies based on gender, race/ethnicity, time since conviction, and type of offense. Methods I use National Longitudinal Survey of Youth 1997 data and restricted comparison group, individual fixed effects, and sibling fixed effects models to examine residential mobility and temporary housing residence during early adulthood. Results I find robust evidence that never-incarcerated individuals with felony convictions experience elevated risk of housing instability and residential mobility, even after adjusting for important mediators like financial resources and relationships. The evidence that incarceration has an additional, independent effect on housing instability is weaker, however, suggesting that the association between incarceration and housing instability found in prior studies may largely be driven by conviction status. Conclusions These findings reveal that conviction, independent of incarceration, introduces instability into the lives of the 12 million Americans who have been convicted of a felony but never imprisoned. Thus, research that attempts to identify an incarceration effect by comparing outcomes to convicted individuals who receive non-custodial sentences may obscure the important independent effect of conviction. Moreover, these findings highlight that the socioeconomic effects of criminal justice contact are broader than incarceration-focused research suggests. Consequently, reform efforts promoting the use of community corrections over incarceration may do less to reduce the harm of criminal justice contact than expected. PubDate: 2022-06-25
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives The paper studies the impact of predictive policing on crime in a developing country. It also assesses the impact of different police trainings. Method We analyze a randomized controlled trial conducted in Montevideo, Uruguay to assess the implementation of a predictive policing software developed in the United States. Half of the precincts were randomly assigned to the software and half to the local crime analysts (status quo). The second experiment allocated randomly a specially trained police force to targeted patrol areas per shift and day. Results No statistically significant differences were found in crime outcomes between the precincts assigned to the foreign predictive software and those assigned to local crime analysts. On the second experiment, given determined targeted places, the specially trained task force showed more compliance with the assigned patrol sites (20% more patrol time) and a greater potential for reducing crime (reduction of 30% in robberies only during high crime shifts in comparison to the control group (no special training). There is also evidence of a diffusion of benefits to adjacent areas. Conclusions The implementation of an international predictive policing software did not outperform local crime analysts in terms of crime reduction. Local crime analysts are more cost-effective. Given determined targeted places, a modest increase in police dosage of a specially trained police force could reduce crime in high-crime times. In developing countries new policing technologies and training require a deep understanding of the context to channel limited resources in the most efficient way. PubDate: 2022-06-24
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives In light of empirical findings suggesting no substantive main effects of an incarcerated person’s (IP’s) race or ethnicity on the odds of placement in restrictive housing (RH) for rule violations, we investigated whether these effects are dependent on offense severity and context, including characteristics of facilities that could theoretically increase stakeholder reliance on biased stereotypes and also prison staff members’ perceptions of danger and order in a facility. Methods Multilevel analyses of race and ethnicity effects on RH decisions, both at the time of the incident (pre-trial) and after the rule infraction hearing, were conducted for all persons admitted to Ohio’s prisons between 2007 and 2016 and found guilty of prison rule violations (N1 = 81,673; N2 = 33). Results We found no significant main effects of an IP’s race or ethnicity on the odds of RH placement for rule infractions, either at the time of the incident or as punishment after a hearing, once the types of violations were controlled. Upon further investigation, we found that African American and Latinx IPs were more likely to receive RH for certain insubordination-related violations, which may invoke greater punitive discretion. Race effects were also stronger in prisons with tighter security, where officers generally relied less on IPs’ acknowledgements of their formal authority for rule enforcement, and in facilities for men. Conclusions Variance in the magnitude of racial and ethnic disparities in the use of RH for rule violations makes sense across prison settings and, as opposed to general race and ethnicity effects, should guide our understanding of the sources of these disparities with the goal of reducing their impacts. PubDate: 2022-06-23
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Brokers are said to be the oiling chain of illicit networks, facilitating the efficient flow of illicit products to destination. Yet, most of the available brokerage measures focus on local or individual networks, missing the brokers who connect others across communities, such as market levels. This study introduces a robust measure that uncovers, scores, and positions these community brokers. Methods We used network data aggregated from numerous investigations related to 1,800 criminal entrepreneurs operating in Western Canada. After uncovering the communities using the Leiden algorithm, we developed a community brokerage score that assesses individual potential reach and control at the meso level, and that accounts for individual position changes due to different community structures. We examined how the score relates to brokerage and structural hole measures as well as seriousness of involvement in criminality. Results We found that the illicit network studied has a strong and stable community structure, and community brokers form about 9% of the population. The score developed is statistically robust and is not strongly related to network and structural hole measures, which confirms the need for a novel measure that captures this strategic position in illicit and other networks. Conclusions Community brokers are especially important in illicit networks where large-scale covert coordination among criminal entrepreneurs is risky. The measure we propose is not overlapping with currently existing brokerage measures and has the potential to contribute to our understanding of how products and information flow beyond local networks, in criminology and other fields. PubDate: 2022-06-18
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Drawing on criminological research about peer delinquency and self-control, we employ a network perspective to identify the potential paths linking impulsivity, peers, and delinquency. We systematically integrate relevant processes into a set of dynamic network models that evaluate these interconnected pathways. Methods Our analyses use data from more than 14,000 students in Pennsylvania and Iowa collected from the evaluation of the PROSPER partnership model. We estimate longitudinal social network models to disentangle the paths through which impulsivity and delinquency are linked in adolescent friendship networks. Results We find evidence of both peer influence and homophilic selection for both impulsivity and delinquency. Further, results indicate that peer impulsivity is linked to individual delinquent behavior through peer influence on delinquency, but not on impulsivity. Finally, the results suggest that impulsivity moderates both influence and selection processes, as adolescents with higher levels of impulsivity are more likely to select delinquent peers but less likely to change their behavior due to peers. Conclusions In sum, this study offers a more holistic framework and stronger theoretical tests than similar studies of the past. Our results illustrate the need to consider the simultaneous network processes related to peers, impulsivity, and delinquency. Further, our findings reveal that a large dataset with ample statistical power is a valuable advantage for detecting the selection processes that shape friendship networks. PubDate: 2022-06-18
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives We reconstruct the networks of officers co-involved in force incidents to test whether interactions with weapon-prone peers impact firearm use. Methods We draw from a statewide dataset of force incidents across law enforcement agencies in New Jersey, and employ conditional likelihood models to estimate whether exposure to peers with histories of firearm use is associated with an officer’s own likelihood of firearm use net of other contextual confounders. Results We find preliminary evidence that officer firearm behaviors, including drawing, pointing, and discharging a firearm, is influenced by an officer’s peers. Greater exposure to colleagues with histories of firearm use is associated with a lower risk of using a firearm. We also find that officer features, including experience and race/ethnicity, are associated with the risk of firearm use. Conclusions Our study suggests officers’ peers structure the risk of firearm use. Our data allow us to look at time order and rule out situational confounders pertaining to firearm use; however, do not allow us to infer causality. We discuss the study’s implications for understanding firearm behaviors and the role of network science in moving policing research forward. PubDate: 2022-06-15
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: A correction to this paper has been published: https://doi.org/10.1007/s10940-021-09511-y PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Area-based prevention studies often produce results that can be represented in a 2-by-2 table of counts. For example, a table may show the crime counts during a 12-month period prior to the intervention compared to a 12-month period during the intervention for a treatment and control area or areas. Studies of this type have used either Cohen’s d or the odds ratio as an effect size index. The former is unsuitable and the latter is a misnomer when used on data of this type. Based on the quasi-Poisson regression model, an incident rate ratio and relative incident rate ratio effect size and associated overdispersion parameter are developed and advocated as the preferred effect size for count-based outcomes in impact evaluations and meta-analyses of such studies. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Mass shootings seemingly lie outside the grasp of explanation and prediction, because they are statistical outliers—in terms of their frequency and severity—within the broader context of crime and violence. Innovative scholarship has developed procedures to estimate the future likelihood of rare catastrophic events such as earthquakes that exceed 7.0 on the Richter scale or terrorist attacks that are similar in magnitude to 9/11. Methods Because the frequency and severity of mass public shootings follow a distribution resembling these previously studied rare catastrophic event classes, we utilized similar procedures to forecast the future severity of these incidents within the United States. Results Using a dataset containing 156 mass public shootings that took place in the U.S. between 1976 and 2018, we forecast the future probability of attacks reaching each of a variety of severity levels in terms of the number of gunfire victims killed and wounded across three different choices of tail model, three different scenarios for future incident rates, and other parameters. Using a set of mid-range parameters, we find that the probability of an event as deadly as the 2017 massacre in Las Vegas occurring before 2040 is 35% (90% uncertainty interval [8, 72]) and we characterize how this projection varies substantially with choice of modeling parameters. Conclusions Our results suggest an uncertain, but concerning, future risk of large-scale mass public shootings, while also illustrating how such forecasts depend on assumptions made about the tail location and other details of the severity distribution model. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Examine whether a death-in-police-custody incident affected community reliance on the police, as measured through citizen calls requesting police assistance for non-criminal caretaking matters. Methods This study used Baltimore Police Department (BPD) incident-level call data (2014–2017) concerning non-criminal caretaking matters (N = 234,781). Counts of non-criminal caretaking calls were aggregated by week for each of 279 unique sections derived from census-tract and police district boundaries. This study devised a Negative Community–Police Relationship Index Score that operationalized the expected risk of a negative community–police relationship for each of the sections. In April 2015, a Baltimore resident, Freddie Gray, died while in BPD custody. A Poisson regression model assessed whether this high-profile death-in-police-custody incident adversely affected the volume of non-criminal caretaking calls to the police and whether that effect was strongest in sections at a high risk of a negative community–police relationship. A falsification test used pocket-dialed emergency calls to verify that any observed trends were not the result of overall telephone usage. Results There was no statistical evidence that the death-in-police-custody incident produced any changes in community reliance on the police for non-criminal caretaking matters, even in high-risk sections. A supplemental analysis using calls for criminal matters yielded similar results. As the falsification test demonstrated, the observed trends were not the result of overall telephone usage. Conclusions Despite a divisive death-in-police-custody incident, citizens were still willing to enlist police assistance. More broadly, the caretaking role of the police may be an important mechanism to strengthen community–police relations, particularly in marginalized neighborhoods vulnerable to strained community–police relations. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Much recent work has focused on how crime concentrates on particular streets within communities. This is the first study to examine how such concentrations vary across the neighborhoods of a city. The analysis evaluates the extent to which neighborhoods have characteristic levels of crime concentration and then tests two hypotheses for explaining these variations: the compositional hypothesis, which posits that neighborhoods whose streets vary in land usage or demographics have corresponding disparities in levels of crime; and the social control hypothesis, which posits that neighborhoods with higher levels of collective efficacy limit crime to fewer streets. Methods We used 911 dispatches from Boston, MA, to map violent crimes across the streets of the city. For each census tract we calculated the concentration of crime across the streets therein using the generalized Gini coefficient and cross-time stability in the locations of hotspots. Results Neighborhoods did have characteristic levels of concentration that were best explained by the compositional hypothesis in the form of demographic and land use diversity. In addition, ethnic heterogeneity predicted higher concentrations of crime over and above what would be expected given the characteristics of the individual streets, suggesting it exacerbated disparities in crime. Conclusions The extent to which crime concentrates represents an underexamined aspect of how crime manifests in each community. It is driven in part by the diversity of places in the neighborhood, but also can be influenced by neighborhood-level processes. Future work should continue to probe the sources and consequences of these variations. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Examines the neighborhood-level relationship between gang graffiti and gang violence in a large city in the western region of the US during a peak period of local gang feuds in 2014–15. Methods Bayesian Poisson log-linear mixed regression models with a spatio-temporal autoregressive process are estimated using a combination of data for N = 42,276 space–time units. Results Consistent with the view of graffiti an important means of street-level communication between gangs and an integral part of group processes associated with violence escalation and contagion, results reveal a roughly 40 to 60% increase in the expected rate of gang homicides, gang assaults, and gang firearm offenses (but not gang robberies) for each unit increase in neighborhood density of gang graffiti. Somewhat unexpectedly, the relationship with both gang homicide and gang assault was stronger for non-threatening gang graffiti than gang graffiti involving explicit threats or disrespect. Conclusions Findings suggest gang graffiti provides clear clues about local “staging grounds,” where gang status is on the line and violence is expected and easily provoked. Thus, while gangs increasingly are dissing rivals and airing beefs through music (e.g., “diss tracks”) and in cyberspace, many still occupy and defend turf and write graffiti that communicates threats to other gangs and feeds into group processes associated with violence escalation and contagion. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives To provide quantitative attention to the correlates of the gender gap in illegal pay. Guided by the literatures on the gendered nature of offending, illegal earnings, and the gender gap in legal pay, we ask: what factors are associated with the gender gap in illegal pay' Methods We use the Delaware Decision Making Study, a sample of incarcerated offenders, to unpack the gender gap in illegal pay with the Blinder-Oaxaca decomposition technique. Results The gender gap in illegal pay is partly accounted for by criminal analogs—criminal capital and psychosocial attributes—to correlates for the gender gap in legal pay and differences in reward structures. Race also emerges as an important factor. Conclusions The disadvantage women face in the legal workforce extends to illegal markets, and our understanding about the gender gap in legal pay can be translated to criminal contexts. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives To test the cumulative disadvantage hypothesis—that system-level racial and ethnic disparities accumulate from intake to final disposition—by investigating relative and absolute disparities across different pathways through the juvenile justice system. Methods Using a sample of 95,670 juvenile court referrals across 140 counties in four states, the present study employed multinomial logistic regression to examine racial and ethnic disparities across 14 possible combinations of juvenile justice outcomes (i.e., pathways), ranked from least to most punitive. We then estimated predicted probabilities and marginal effects of race and ethnicity for each pathway. Results We found limited support for the cumulative disadvantage hypothesis. Racial and ethnic disparities were greatest for the most punitive pathways, but the findings do not point to extensive evidence of cumulative disadvantage. Specifically, neither relative nor absolute disparities accumulated from least to most punitive pathways, and some of the least punitive pathways were actually more likely for minority defendants. Conclusions The results underscore the need for more careful measurement and analysis of disadvantage and disparities in the criminal and juvenile justice systems. In particular, more attention should be paid to early outcomes such as detention, where large differences between racial and ethnic groups were observed, as well as to relative and absolute differences in processing outcomes. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives To examine how the practice of daily proactivity affects and responds to changes in crime at micro geographic and temporal scales. Methods Police calls for service and automated vehicle location data from a large suburban jurisdiction were used to create comprehensive measures of police proactivity. Panel data and the generalized method of moments framework were applied to tease out the endogenous relationships between crime and police proactivity and understand the unique impact of proactive patrol and crime upon one another. Results Daily police proactivity in this locality was highly stable at micro places, although police did intensify their activities very briefly in response to recent changes in crime. In turn, increases in proactive patrol generated immediate increases in crime reporting, followed by fleeting residual deterrent effects that were weaker and less robust. The patterns remained relatively consistent when varying the units of analysis or focusing on hot spots with different profiles of proactivity, but the deterrent effects appeared more sensitive to model specification. Of all measures of proactivity, patrols of medium length and non-traffic enforcement activities were associated with stronger evidence of crime reduction effects. Conclusions Short-term adjustments in hot spot patrols appear to produce both reporting effects and temporary residual deterrent effects as measured through calls for service and police vehicle location data. Police could potentially enhance and prolong their deterrence by adopting more deliberate strategies with their daily proactive behaviors, including making their proactive activities more targeted and sustained. PubDate: 2022-06-01
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objective This study examines the mechanisms underpinning the emergence of violence among individuals in the organized crime milieu. Methods Relying on criminal event data recorded by a UK Police Force, we apply a longitudinal network approach to study violent interactions among offenders. The data span the period from 2000 to 2016 and include 6,234 offenders and 23,513 organized crime-related events. Instead of aggregating these data over time, we use a relational event-based approach to take into consideration the order of events. We employ an actor-oriented framework to model offenders’ victim choices in 156 violent events in the OC milieu. Results We find that the choice of offenders to target a particular victim is strongly affected by their mutual history. A violent act is often preceded by a previous act of violence, both in the form of repeated violence and reciprocated violence. We show that violence is strongly associated with prior co-offending turning sour. We uncover a strong effect for previous harassment as a retaliation cum escalation mechanism. Finally, we find evidence of conflicts within organized crime groups and of violence being directed to offenders with the same ethnic background. Conclusions Relational effects on victimization are consistently stronger than the effects of individual characteristics. Therefore, from a policy perspective, we believe that relational red flags (or risk factors) should play a more central role. A focus on harassment could be valuable in the development of an early intervention strategy. PubDate: 2022-04-12 DOI: 10.1007/s10940-022-09540-1
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives Scholars and practitioners have paid increasing attention to problematic properties, but little is known about how they emerge and evolve. We examine four phenomena suggested by life-course theory that reflect stability and change in crime and disorder at properties: onset of issues; persistence of issues; aggravation to more serious types of issues; and desistance of issues. We sought to identify the frequency and dynamics of each. Methods We analyze how residential parcels (similar to properties) in Boston, MA shifted between profiles of crime and disorder from 2011 to 2018. 911 dispatches and 311 requests provided six measures of physical disorder, social disorder, and violence for all parcels. K-means clustering placed each parcel into one of six profiles of crime and disorder for each year. Markov chains quantified how properties moved between profiles year-to-year. Results Onset was relatively infrequent and more often manifested as disorder than violence. Pathways of aggravation led from less serious profiles to a mixture of violence and disorder. Desistance was more likely to occur as de-escalations along these pathways then complete cessation of issues. In neighborhoods with above-average crime, persistence was more prevalent whereas desistance less often culminated in cessation, even relative to local expectations. Conclusions The results offer insights for further research and practice attentive to trends of crime and disorder at problematic properties. It especially speaks to the understanding of stability and change; the role of different types of disorder; and the toolkit needed for problem properties interventions. PubDate: 2022-03-30 DOI: 10.1007/s10940-022-09542-z
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives This paper estimates the effect of changes in street lighting at night on levels of crime at street-level. Analyses investigate spatial and temporal displacement of crime into adjacent streets. Methods Offense data (burglaries, robberies, theft of and theft from vehicles, and violent crime) were obtained from Thames Valley Police, UK. Street lighting data (switching lights off at midnight, dimming, and white light) were obtained from local authorities. Monthly counts of crime at street-level were analyzed using a conditional fixed-effects Poisson regression model, adjusting for seasonal and temporal variation. Two sets of models analyzed: (1) changes in night-time crimes adjusting for changes in day-time crimes and (2) changes in crimes at all times of the day. Results Switching lights off at midnight was strongly associated with a reduction in night-time theft from vehicles relative to daytime (rate ratio RR 0.56; 0.41–0.78). Adjusted for changes in daytime, night-time theft from vehicles increased (RR 1.55; 1.14–2.11) in adjacent roads where street lighting remained unchanged. Conclusion Theft from vehicle offenses reduced in streets where street lighting was switched off at midnight but may have been displaced to better-lit adjacent streets. Relative to daytime, night-time theft from vehicle offenses reduced in streets with dimming while theft from vehicles at all times of the day increased, thus suggesting temporal displacement. These findings suggest that the absence of street lighting may prevent theft from vehicles, but there is a danger of offenses being temporally or spatially displaced. PubDate: 2022-03-30 DOI: 10.1007/s10940-022-09539-8
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives To provide a detailed understanding of how the prevalence and frequency of offending vary with age in the Cambridge Study in Delinquent Development (CSDD) and to quantify the influence of early childhood risk factors such as high troublesomeness on this variation. Methods We develop a statistical model for the prevalence and frequency of offending based on the hurdle model and curves called splines that allow smooth variation with age. We use the Bayesian framework to quantify estimation uncertainty. We also test a model that assumes that frequency is constant across all ages. Results For 346 males from the CSDD for whom the number of offenses at all ages from 10 to 61 are recorded, we found peaks in the prevalence of offending around ages 16 to 18. Whilst there were strong differences in prevalence between males of high troublesomeness and those of lower troublesomeness up to age 45, the level of troublesomeness had a weaker effect on the frequency of offenses, and this lasted only up to age 20. The risk factors of low nonverbal IQ, poor parental supervision and low family income affect how prevalence varies with age in a similar way, but their influence on the variation of frequency with age is considerably weaker. We also provide examples of quantifying the uncertainty associated with estimates of interesting quantities such as variations in offending prevalence across levels of troublesomeness. Conclusions Our methodology provides a quantified understanding of the effects of risk factors on age-crime curves. Our visualizations allow these to be easily presented and interpreted. PubDate: 2022-03-29 DOI: 10.1007/s10940-022-09544-x
Please help us test our new pre-print finding feature by giving the pre-print link a rating. A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Objectives We study interpretable recidivism prediction using machine learning (ML) models and analyze performance in terms of prediction ability, sparsity, and fairness. Unlike previous works, this study trains interpretable models that output probabilities rather than binary predictions, and uses quantitative fairness definitions to assess the models. This study also examines whether models can generalize across geographic locations. Methods We generated black-box and interpretable ML models on two different criminal recidivism datasets from Florida and Kentucky. We compared predictive performance and fairness of these models against two methods that are currently used in the justice system to predict pretrial recidivism: the Arnold PSA and COMPAS. We evaluated predictive performance of all models on predicting six different types of crime over two time spans. Results Several interpretable ML models can predict recidivism as well as black-box ML models and are more accurate than COMPAS or the Arnold PSA. These models are potentially useful in practice. Similar to the Arnold PSA, some of these interpretable models can be written down as a simple table. Others can be displayed using a set of visualizations. Our geographic analysis indicates that ML models should be trained separately for separate locations and updated over time. We also present a fairness analysis for the interpretable models. Conclusions Interpretable ML models can perform just as well as non-interpretable methods and currently-used risk assessment scales, in terms of both prediction accuracy and fairness. ML models might be more accurate when trained separately for distinct locations and kept up-to-date. PubDate: 2022-03-28 DOI: 10.1007/s10940-022-09545-w
Hybrid journal (It can contain Open Access articles)
[2469 journals]
