Article
To read the full-text of this research, you can request a copy directly from the authors.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... To make binary, forced-choice linkage decisions, 1 any crime pair associated with a probability value above a pre-specified threshold is predicted to be linked. 2 While LR analysis is often associated with moderate to high AUCs when used for this purpose (Bennell et al. 2014), it typically results in nomothetic linking strategies (i.e. the same predictor variables are applied the same way in every case) that will miss potential links (Tonkin et al. 2012b). For example, a LR model might always result in a prediction that two crimes are linked when the distance between those crimes is small (e.g. ...
... Lundrigan et al. 2010). Concerns regarding this "one-size-fits-all" approach has prompted research into other statistical crime linkage methods that might be more useful for capturing the heterogeneity of serial offender behaviour (Tonkin et al. 2012b). ...
... For example, Tonkin et al. (2012b) compared linking models produced through LR analysis and classification tree (CT) analysis. As Tonkin and his colleagues discuss, a CT consists of a structured set of questions (related to specific predictor variables) that can be used by practitioners to systemically decide whether crime pairs have been committed by the same offender (e.g. is the distance between these two crimes > or < 2.4 km?). ...
Article
Full-text available
Studies have shown that it is possible to link serial crimes in an accurate fashion based on the statistical analysis of crime scene information. Logistic regression (LR) is one of the most common statistical methods in use and yields relatively accurate linking decisions. However, some research suggests there may be added value in using classification tree (CT) analysis to discriminate between offences committed by the same vs. different offenders. This study explored how three variations of CT analysis can be applied to the crime linkage task. Drawing on a sample of serial sexual assaults from Quebec, Canada, we examine the predictive accuracy of standard, iterative, and multiple CTs, and we contrast the results with LR analysis. Our results revealed that all statistical approaches achieved relatively high (and similar) levels of predictive accuracy, but CTs produce idiographic linking strategies that may be more appealing to practitioners. Future research will need to examine if and how these CTs can be useful as decision aides in operational settings.
... Melnyk et al. (2011) reported two studies, one examining serial homicide and one examining serial burglary. Because Tonkin, Woodhams, Bull, Bond, & Santtila (2012) also examined two different crime types from different locations, their paper was also classified as containing two separate studies for the purposes of our review. ...
... The AUC ranges in studies of car theft are also very wide (0.54–0.93 in Davies et al., 2012; 0.56–0.81 in Tonkin et al., 2008; and 0.50–0.82 in Tonkin, Woodhams, Bull, Bond, & Santtila, 2012), as is the range in the lone study of personal robbery (0.45–0.92; Burrell et al., 2012). ...
... Upon examination of the studies in the Appendix, there seems to be relatively consistent variation from one behavioural domain to the next in terms of the AUC. The most consistent finding is that inter-crime distance and temporal proximity are associated with some of the largest AUC values, and these values often exceed the AUCs associated with more common MO behaviours (such as the type of home that was targeted in a burglary or what property was stolen; Bennell & Canter, 2002; Bennell & Jones, 2005; Markson et al., 2010; Tonkin, Woodhams, Bull, Bond, & Santtila, 2012). As Tonkin (forthcoming) has noted, similar findings have also been observed with car theft (Davies et al., 2012; Tonkin et al., 2008) and personal robbery (Burrell et al., 2012), but research examining the potential value of spatial and temporal variables in cases of sexual assault/rape and homicide is lacking. ...
Article
The number of published studies examining crime linkage analysis has grown rapidly over the last decade, to the point where a special issue of this journal has recently been dedicated to the topic. Many of these studies have used a particular measure (the area under the receiver operating characteristic curve, or the AUC) to quantify the degree to which it is possible to link crimes. This article reviews studies that have utilised the AUC and examines how good we are currently at linking crimes (within the context of these research studies) and what factors impact linking accuracy. The results of the review suggest that, in the majority of cases, moderate levels of linking accuracy are achieved. Of the various factors that have been examined that might impact linking accuracy, the three factors that appear to have the most significant impact are crime type, behavioural domain, and jurisdiction. We discuss how generalisable these results are to naturalistic investigative settings. We also highlight some of the important limitations of the linking studies that we reviewed and offer up some strategies for moving this area of research forward.
... Prediction of the class label of an instance is done by tracing the decision tree starting from the root node to the leaf node. The decision tree model can handle mixed variables and has a high degree of accuracy [39]. Decision tree construction is top-down, which is done by partitioning the set of instances recursively by selecting an attribute that has the ability to separate the set of instances with the highest separation at each partitioning process [40]. ...
... According to the results by Tonkin et al. [39] and by Blanquero et al. [40] stated that the performance of the decision tree model has outperformed the performance of the logistic regression model. The research has contradictory results where the decision tree model performance is lower than the logistic regression performance. ...
Article
Full-text available
In everyday life, we are always faced with making decisions to choose the right decision between 2 choices of decision candidates. The development of high-performance binary classification models is a challenge for researchers in the modeling field. Deployment both of logistic regression and decision tree model use the dataset having predictor features which are a mixture of categorical and numerical features, both models tend to suffer an overfitting problem. This study has the aim of building a ridge logistic regression and decision tree model on a dataset that has all features of a binary categorical scale. The novelty of this study is to observe the distribution of the two classes in the dataset using the transformation of principal components and linear discriminant projections and also to explore the importance of feature that plays a role in building the decision tree model. The ridge logistic regression model has an accuracy performance of 84% which is better than the decision tree model having an accuracy performance of 81%. There are only 2 features in the dataset dominating around 80% of the feature importance.
... (Krzanowski & Hand, 2009 Markson et al., 2010;Tonkin et al., 2012aTonkin et al., , 2012c) ,強盗 (Burrell et al., 2012;Woodhams & Toye, 2007) ,自動車盗またはエンジン キ ー の 窃 盗 (Davies et al., 2012;Tonkin et al., 2008Tonkin et al., , 2012c)といった犯罪について地理的・時間的近接性 による事件リンクを検討した先行研究と同等のレベル である。したがって,本研究では,英国において罪種 横断的な事件リンクの有効性を示した研究 (Tonkin et al., 2011(Tonkin et al., , 2012b (Bernasco, 2009;Block & Bernasco, 2009;Emeno & Bennell, 2013;Haginoya, 2014;Hammond & Youngs, 2011;Laukkanen, Santtila, Jern, & Sandnabba, 2008;Rattner & Portnov, 2007;Rengert, Piquero, & Jones, 1999;Rhodes & Conly, 1981;Sarangi & Youngs, 2006;Snook, 2004 2002;Bennell & Jones, 2005;Burrell et al., 2012;萩野谷, 2014;Markson et al., 2010;Tonkin et al., 2011;Woodhams, 2008 ...
... ) ,強盗 (Burrell et al., 2012;Woodhams & Toye, 2007) ,自動車盗またはエンジン キ ー の 窃 盗 (Davies et al., 2012;Tonkin et al., 2008Tonkin et al., , 2012c)といった犯罪について地理的・時間的近接性 による事件リンクを検討した先行研究と同等のレベル である。したがって,本研究では,英国において罪種 横断的な事件リンクの有効性を示した研究 (Tonkin et al., 2011(Tonkin et al., , 2012b (Bernasco, 2009;Block & Bernasco, 2009;Emeno & Bennell, 2013;Haginoya, 2014;Hammond & Youngs, 2011;Laukkanen, Santtila, Jern, & Sandnabba, 2008;Rattner & Portnov, 2007;Rengert, Piquero, & Jones, 1999;Rhodes & Conly, 1981;Sarangi & Youngs, 2006;Snook, 2004 2002;Bennell & Jones, 2005;Burrell et al., 2012;萩野谷, 2014;Markson et al., 2010;Tonkin et al., 2011;Woodhams, 2008 ...
Article
Full-text available
A number of studies have investigated behavioral case linkage between crimes of a specified type, such as linking one residential burglary with another residential burglary. However, only a few studies have investigated the effectiveness of case linkage across crime types, which have been limited to the UK. This study examined whether linking across crime types using spatio-temporal proximity was possible in samples that were different from the UK in terms of the structure of crime classification. This was accomplished by calculating the discrimination accuracy between linked crime pairs (two offenses committed by the same offender) and unlinked crime pairs (two offenses committed by different offenders) using geographical (inter-crime distance) and temporal proximity (number of days between offenses) across various crimes committed in Japan. Both the geographical proximity and temporal proximity had statistically significant levels of discrimination accuracy across crime types as assessed by Receiver Operating Characteristic (ROC) analysis. This suggests the possibility of identifying a crime series by geographical and temporal proximity across multiple crime types in Japan.
... This is a binary classification problem where each crime pair is considered independently. There have been numerous studies detailing the performance of case linkage methods across a variety of crime types (e.g., Bennell and Jones, 2005;Bennell et al., 2009;Brown and Hagen, 2003;Cocx and Kosters, 2006;Goodwill and Alison, 2006;Lin and Brown, 2006;Markson et al., 2010;Tonkin et al., 2008Tonkin et al., , 2011Tonkin et al., , 2012aWoodhams and Labuschagne, 2012;Woodhams and Toye, 2007). ...
... To facilitate complex interactions between the evidence variables, classification trees (Tonkin et al., 2012a), random forests, gradient boosting, and other ensemble methods can be employed (Berk, 2006(Berk, , 2013. ...
Article
Full-text available
The object of this paper is to develop a statistical approach to criminal linkage analysis that discovers and groups crime events that share a common offender and prioritizes suspects for further investigation. Bayes factors are used to describe the strength of evidence that two crimes are linked. Using concepts from agglomerative hierarchical clustering, the Bayes factors for crime pairs are combined to provide similarity measures for comparing two crime series. This facilitates crime series clustering, crime series identification, and suspect prioritization. The ability of our models to make correct linkages and predictions is demonstrated under a variety of real-world scenarios with a large number of solved and unsolved breaking and entering crimes. For example, a naive Bayes model for pairwise case linkage can identify 82% of actual linkages with a 5% false positive rate. For crime series identification, 74%-89% of the additional crimes in a crime series can be identified from a ranked list of 50 incidents.
... Por ejemplo, utilizaron esta técnica GROSSI et al. (2000) o RODRÍGUEZ DÍAZ, PAÍNO y de la VILLA (2007).13 Algunos estudios han comparado la precisión de esta herramienta estadística con la de otras técnicas, entre ellas la regresión logística, obteniendo, en general, resultados equivalentes o, incluso, superiores para la primera, si bien cada técnica presenta un conjunto específico de fortalezas y áreas de mejora(véanse, por ejemplo, STALANS et al., 2004;ROSENFELD y LEWIS, 2005;YANG, LIU y COID, 2010;LIU et al., 2011;TONKIN et al., 2012;NGO, GOVINDU y AGARWAL, 2015;TONKIN et al., 2017). ...
Article
Full-text available
En el presente estudio se analizaron las relaciones entre tres conjuntos de variables: características sociodemográficas, condiciones de vida y antecedentes personales y familiares, y las experiencias de victimización en una muestra de personas en situación de sin hogar. El objetivo fue determinar qué variables se encontraban en mayor medida relacionadas con dichas experiencias para este colectivo. Aplicando análisis de segmentación y, posteriormente, análisis de regresión logística, se obtuvo que las experiencias de victimización se encontraban conectadas con una combinación de condiciones de vida desfavorables y antecedentes adversos, con resultados esencialmente convergentes entre ambas estrategias analíticas. Sin embargo, el análisis de segmentación permitió además identificar de forma automática algunas interacciones relevantes entre las variables. Por ejemplo, para aquellos que habían sido detenidos en dependencias policiales en más de una ocasión, el mejor pronosticador de las experiencias de victimización fue contar con antecedentes de problemas de alcoholismo en la familia o de la propia persona durante la infancia o adolescencia, mientras que para quienes sólo habían sido detenidos una vez la variable más relevante fue el tiempo que llevaban en situación de sin hogar. Entre los participantes que no habían sido detenidos, el factor más relacionado con la victimización fue el estado de salud percibido. // In this paper, relationships among three sets of variables: sociodemographic characteristics, living conditions and family history and personal background, and criminal victimisation events in a sample of homeless people were analysed. The aim of the study was to identify which variables were related to those experiences to a greater extent. By applying segmentation analysis and, afterwards, logistic regression, it was found that criminal victimisation experiences were connected to a combination of different unfavourable living conditions and an adverse background, with essentially converging findings between both analytic strategies. However, segmentation analysis also automatically provided evidence of some relevant interactions among variables. For instance, the best predictor of victimisation events among those who had been arrested more than once were personal or family problems with alcohol during childhood or adolescence, Title: Criminal victimisation events of people who are homeless. An approach to victim's profile by means of segmentation analysis and logistic regression.-Palabras clave: Personas sin hogar, victimización, condiciones de vida, antecedentes personales, antecedentes familiares, análisis CHAID, regresión logística.
... There are several offender risk assessment applied studies that show that the classifying performance between DT and LR is similar in different classification problems: Steadman et al. (2000) compared the Chi-Squared Automatic Interaction Detector (CHAID) and LR to develop violence risk assessment models; S. Thomas et al. (2005) compared the performance of the Classification And Regression Tree (CART) decision algorithm and LR to develop models to predict violence in psychotic illness; Rosenfeld and Lewis (2005) compared the CART decision algorithm and LR regarding the problem of assessing violence risk in stalking cases; R. Berk, Sherman, Barnes, Kurtz, and Ahlman (2009) compared random forest and LR to construct a predictive model of murder within a population of probationers and parolees; Liu, Yang, Ramsay, Li, and Coid (2011) compared the CART, neural network and LR in predicting violent re-offending; Neuilly, Zgoba, Tita, and Lee (2011) compared classification tree analysis, random forest, and LR to predict recidivism in homicide offenders; another important work was conducted by Hamilton, Neuilly, Lee, and Barnoski (2014), who suggested that machine-learning techniques, including DT algorithms, have a classifying performance similar to LR in offender risk assessment. Finally, the results of the present study should be seen as complementary to those of an earlier work (Tonkin et al., 2012), in which the relative performances of DT and LR were compared for the case linkage problem. ...
Article
Full-text available
The aim of this paper was to create a decision tree (DT) to identify personality profiles of offenders against public safety. A technique meeting this requirement was proposed that uses the C4.5 algorithm to derive decision rules for personality profiling of public safety offenders. The Mini-Mult test was used to measure the personality profiles of 238 individuals. With the test results as our database, a C4.5 DT was applied to construct rules that classify each profile into one of two groups, those without and those with records of offences against public safety. The model correctly classified 80% of the personality profiles and delivered a set of decision rules for distinguishing the profiles by group, and the principal personality profiles were interpreted. We conclude that DTs are a promising technique for analysing personality profiles by their offender or non-offender status. Finally, we believe that the development of a classifying model using DT may have practical applications in the Colombian prison system.
... As such, the main emphasis in case linkage research has been in determining how well the assumptions regarding consistency and distinctiveness hold. This has resulted in numerous studies detailing the performance of several case linkage methods across a variety of crime types (Brown and Hagen, 2003;Bennell and Jones, 2005;Goodwill and Alison, 2006;Lin and Brown, 2006;Cocx and Kosters, 2006;Woodhams and Toye, 2007;Tonkin et al., 2008;Bennell et al., 2009;Markson et al., 2010;Tonkin et al., 2011Tonkin et al., , 2012Woodhams and Labuschagne, 2012). ...
Article
Statistical clustering of criminal events can be used by crime analysts to create lists of potential suspects for an unsolved crime, to identify groups of crimes that may have been committed by the same individuals or group of individuals, for offender profiling and for predicting future events. We propose a Bayesian model-based clustering approach for criminal events. Our approach is semisupervised, because the offender is known for a subset of the events, and utilizes spatiotemporal crime locations as well as crime features describing the offender's modus operandi. The hierarchical model naturally handles complex features that are often seen in crime data, including missing data, interval-censored event times and a mix of discrete and continuous variables. In addition, our Bayesian model produces posterior clustering probabilities which allow analysts to act on model output only as warranted. We illustrate the approach by using a large data set of burglaries in 2009–2010 in Baltimore County, Maryland.
Article
Full-text available
In many countries, 40–60% of the traffic accidents occur at junctions, making the reduction of junction accidents paramount to achieving UN Sustainable Development Goals. In Japan, the road safety guidelines specify the proximity between junctions and non-perpendicular angles at junctions as the two main risk factors behind junction accidents, yet their impact remains understudied. Using binomial logistic regression models, this study investigates the impact of junction intervals and junction angles on the severity of traffic accidents. The study found that, in general, (1) shorter intervals between adjacent junctions helps reduce the risk of serious accidents, which is the opposite of the current road safety guidelines in Japan, and (2) results from the junction angle analysis were mixed but there was no evidence that the roads should meet at a right angle to reduce traffic accidents. Some types of accidents also returned a non-linear curve, e.g., vehicle-to-vehicle collisions at four-armed junctions involving a driver aged 65 years and over have the highest risk of fatal/serious accidents when adjacent junctions were 32 m apart, and the risk reduces at a shorter or longer interval. These results suggest that the current road safety guidelines require updating to improve road safety around junctions.
Article
Full-text available
The growth and development of children under five (toddlers) affect their health conditions. Each region uniquely identifies the main factors influencing the toddler's health condition. The status of toddlers is generally categorized into two classes, namely normal and abnormal, so it is often found that the condition of toddler status is in the form of multi-response variables. Combining the two binary classes' response variables will form a multiclass response variable requiring different model development techniques and performance measurements. This study aims to determine the main factors that affect toddlers' health conditions in Malang, Indonesia, build multiclass logistic regression and decision tree classification models, and measure the model's performance. The Chi-square test selected predictor features as the input of multiclass logistic regression and decision tree models. From the feature selection, four main factors influence the status of toddlers' health conditions in Malang: the mother's history of diabetes before pregnancy, the father's blood pressure, psychological condition, and drinking water quality. The decision tree model performs better than the logistic regression model on the various performance measures used.
Chapter
Research has shown that the majority of offences are committed by a minority of offenders. Therefore, any method to help identify prolific/serial offenders is of benefit to the police. Behavioural Crime Linkage (BCL) is a method of identifying series of offences committed by the same person(s) using the behaviour displayed during the offence. This can include, but is not limited to, target selection, control and weapon use, approach, property stolen, and temporal and spatial trends. This chapter will explain the theoretical framework for BCL and common methods for testing the accuracy of this method (e.g. logistic regression, Receiver Operating Characteristic ). The chapter will then outline how BCL has been applied in robbery. It will discuss how the success of BCL is influenced by factors such as type of location (e.g. urban versus rural) and group offending (e.g. can you link offences committed by groups?). This chapter will draw heavily on the PhD research of the author but will cite other literature (e.g. evidence to support the theoretical framework for BCL) where relevant.KeywordsBehavioural crime linkageCrime linkageRobbery
Article
Crime linkage is a difficult task and is of great significance to maintaining social security. It can be treated as a binary classification problem. Some crimes are difficult to determine whether they are serial crimes under the existing evidence, so the two-way decisions are easy to make mistakes for some case pairs. Here, the three-way decisions based on the decision-theoretic rough set are applied and its key issue is to determine thresholds by setting appropriate loss functions. However, sometimes the loss functions are difficult to obtain. In this paper, a method to automatically learn thresholds of the three-way decisions without the need to preset explicit loss functions is proposed. We simplify the loss function matrix according to the characteristic of crime linkage, re-express thresholds by loss functions, and investigate the relationship between overall decision cost and the size of the boundary region. The trade-off between the uncertainty of the boundary region and the decision cost is taken as the optimization objective. We apply multiple traditional classification algorithms as base classifiers, and employ real-world cases and some public datasets to evaluate the effect of our proposed method. The results show that the proposed method can reduce classification errors.
Article
Full-text available
Detecting serial crimes is to find criminals who have committed multiple crimes. A classification technique is often used to process serial crime detection, but the pairwise comparison of crimes is of quadratic complexity, and the number of nonserial case pairs far exceeds the number of serial case pairs. The blocking method can play a role in reducing pairwise calculation and eliminating nonserial case pairs. But the limitation of previous studies is that most of them use a single criterion to select blocks, which is difficult to guarantee an excellent blocking result. Some studies integrate multiple criteria into one comprehensive index. However, the performance is easily affected by the weighting method. In this paper, we propose a combined blocking (CB) approach. Each criminal behaviour is defined as a behaviour key (BHK) and used to form a block. CB learns several weak blocking schemes by different blocking criteria and then combines them to form the final blocking scheme. The final blocking scheme consists of several BHKs. Because rare behaviour can better identify crime series, each BHK is assigned a score according to its rarity. BHKs and their scores are used to determine whether a case pair need to be compared. After comparing with multiple blocking methods, CB can effectively guarantee the number of serial case pairs while greatly reducing unnecessary nonserial case pairs. The CB is embedded in a supervised machine learning framework. Experiments on real-world robbery cases demonstrate that it can effectively reduce pairwise comparison, alleviate the class imbalance problem and improve detection performance.
Article
Crime linkage is a challenging task in crime analysis, which is to find serial crimes committed by the same offenders. It can be regarded as a binary classification task detecting serial case pairs. However, most case pairs in the real world are nonserial, so there is a serious class imbalance in the crime linkage. In this paper, we propose a novel random forest based on the information granule. The approach doesn’t resample the minority class or the majority class but concentrates on indistinguishable case pairs at the classification boundary. The information granule is used to identify case pairs that are difficult to distinguish in the dataset and constructs a nearly balanced dataset in the uncertainty region to deal with the imbalanced problem. In the proposed approach, random trees come from the original dataset and the above mentioned nearly balanced dataset. A real-world robbery dataset and some public imbalanced datasets are employed to measure the performance of the approach. The results show that the proposed approach is effective in dealing with class imbalances, and it can be extended to combine with other methods solving class imbalances.
Article
Full-text available
Whistleblower lawsuits under the federal False Claims Act have markedly increased over the past decade. While the amount of individual settlements and judgments vary, over time the government has paid out an average of about 15% of its recoveries as awards to whistleblowers. This investigation used a classification tree algorithm to analyze a sample of recent False Claims Act qui tam settlements, identifying several factors that distinguish larger settlements from smaller ones. Notably, the public or private status of corporate defendants, the federal judicial circuit in which the case is settled, type of case, case duration, relator status, and whether the Department of Justice intervened in the case are significant indicators of settlement amount in qui tam cases. Companies can use this information to better evaluate their exposure to liability for whistleblower litigation, while potential whistleblowers and their counsel can use it to evaluate the likelihood of winning an award.
Article
Full-text available
This study compared the ability of seven statistical models to distinguish between linked and unlinked crimes. The seven models utilised geographical, temporal, and modus operandi information relating to residential burglaries (n = 180), commercial robberies, (n = 118), and car thefts (n = 376). Model performance was assessed using receiver operating characteristic analysis and by examining the success with which the seven models could successfully prioritise linked over unlinked crimes. The regression‐based and probabilistic models achieved comparable accuracy and were generally more accurate than the tree‐based models tested in this study. The Logistic algorithm achieved the highest area under the curve (AUC) for residential burglary (AUC = 0.903) and commercial robbery (AUC = 0.830) and the SimpleLogistic algorithm achieving the highest for car theft (AUC = 0.820). The findings also indicated that discrimination accuracy is maximised (in some situations) if behavioural domains are utilised rather than individual crime scene behaviours and that the AUC should not be used as the sole measure of accuracy in behavioural crime linkage research.
Article
Full-text available
In the 4 decades since offender profiling (OP) was established, hundreds of journal articles, books, book chapters, reports, and magazine articles have been published on the topic, and the technique has been used by countless law enforcement agencies around the globe. However, despite the popularity and extensive literature published on OP, very little is known about its evolution, current state, or findings of the field to date. Therefore, this study presents a systematic review and meta-analysis of 426 publications on OP from 1976 through 2016. Results of this systematic review suggest that there have been considerable improvements in the scientific rigor and self-assessment being conducted in the discipline, although in total, few studies have used a strong empirical approach to develop new profiles. Even fewer evaluations of the effectiveness of OP have been conducted. The first summary of offender profiles proposed for major crimes in OP literature is also presented, with results indicating some recurrent themes in profiles, but wide variations in the number, name, and description of the profiles often found. A meta-analysis of case linkage analysis research indicates that this area is statistically sophisticated, and has yielded moderate to strong accuracy rates for linking crimes to a single offender. Finally, the first analysis of the most prolific authors, researchers, departments, and outlets for OP research, and the methods, approach, and most cited publications in OP are identified. Suggestions for future research on OP and the potential impact that this may have on policy and practice are also discussed. (PsycINFO Database Record (c) 2018 APA, all rights reserved).
Article
The present study compared decision tree analysis to logistic regression analysis in order to investigate whether decision tree analysis has sufficient ability to construct a model that predicts offender characteristics from the crime scene and/or victim information. The data used in this study were collected from solved single homicide cases that occurred in Japan between 2004 and 2009 (n=1226). After constructing models that predict offender's criminal history by logistic regression analysis and decision tree analysis, AUC (area under the ROC curve) of those models and the predictive values were compared. The AUC was .75 (p<.001) for logistic regression model and .71 (p<.001) for the decision tree model. A significant difference between these AUCs was not observed (χ²(1)=3.71, p=.05). The predictive values were 67.3% for both the logistic regression model and the decision tree model. These findings suggest that the decision tree is comparable to logistic regression analysis in constructing a model that predicts the offender's criminal history from offence characteristics in single homicide cases.
Article
Full-text available
Patent litigation is expensive—very expensive. Obviously, the high cost of patent litigation makes the patent litigation game a high stakes one. The possibility of facing an alleged patent infringer or being sued by a patent holder can be daunting given the aforementioned high costs involved and the possibility of losing the value of the patent in the lawsuit from a finding of invalidity or unenforceability. Thus, the ability to predict the outcome of patent litigation would allow parties to gauge the potential risks and costs and make a more informed decision about how to proceed when facing a patent lawsuit. Though a number of studies have used standard logistic regression models using patent characteristics to study patent litigation, and at least one study used case facts to predict patent litigation outcomes, we add the use of classification trees to find predictors in patent litigation outcomes. Both models are valid and useful ways to examine and predict patent litigation. We propose that using classification trees as a tool to predict patent litigation may be easier than using the logistic regression model, which traditionally has a more mathematical interpretation. We conclude that that classification tree analysis is as good as logistic regression analysis in predicting win/loss ratios and provides richer information about the hierarchical nature of patent case decisions and the interaction of variables that influence those decisions.
Article
Full-text available
This study aimed to investigate the case linkage principles, behavioural consistency and distinctiveness, with a sample of serial car thieves. Target selection, acquisition, and disposal behaviours, as well as geographical and temporal behaviours, were examined. The effects of temporal proximity and offender expertise were also investigated as moderating factors of behavioural consistency. As in previous case linkage research, geographical and some target selection behaviours were able to predict whether crime pairs are linked or unlinked at a statistically significant level. Crucially, it was also found that temporal behaviours demonstrate a significant capability to predict linkage status, a variable which has never before been applied to the prediction of linkage in serial car theft. Furthermore, it was demonstrated that changing the operationalisation of the behavioural domains can affect the results obtained. No support was found for the moderation of behavioural consistency on the basis of temporal proximity or expertise. Overall, the results support previous case linkage studies, furthering their practical applicability within the criminal justice system. Copyright © 2012 John Wiley & Sons, Ltd.
Article
Full-text available
The aim of the present study was to predict offender characteristics on the basis of crime scene behaviour in urban burglaries. The police files of 633 burglaries in the Finnish Metropolitan Area between 1990 and 2001 were content analysed using a predetermined list of variables. The crime scene behaviour variables were subjected to a principal component analysis. Fourteen factors indicative of different types of burglaries were identified and used to predict the characteristics of the 244 offenders using regression models. Statistically significant predictors of almost all offender characteristics were identified. From a practical point of view, the predictive models could be used in police investigations to narrow down the number of suspects.
Article
Full-text available
The assigning of different weights to risk factors in actuarial formulas for the assessment of violence risk in criminal offenders has been debated. The authors explore the predictive validity of an index with 10 well-established risk factors for criminal recidivism with respect to violent reconvictions among 404 former forensic psychiatric examinees in Sweden. Four different weighting conditions are tested experimentally, including Nuffield’s method, bivariate and multivariate logistic regression, and an artificial neural network procedure. Simpler weighting techniques do not improve predictive accuracy over that of a nonweighted reference, and the more complex procedures yield a statistical shrinkage effect. The authors hypothesize that the general lack of causal risk factors in prediction models may contribute to the observed low utility of weighting techniques.
Article
Full-text available
The extrapolation of characteristics of criminals from information about their crimes, as an aid to police investigation, is the essence of ‘profiling'. This paper proposes that for such extrapolations to be more than educated guesses they must be based upon knowledge of (1) coherent consistencies in criminal behaviour and (2) the relationship those behavioural consistencies have to aspects of an offender available to the police in an investigation. Hypotheses concerning behavioural consistencies are drawn from the diverse literature on sexual offences and a study is described of 66 sexual assaults committed by 27 offenders against strangers. Multivariate statistical analyses of these assaults support a five-component system of rapist behaviour, reflecting modes of interaction with the victim as a sexual object. The potential this provides for an eclectic theoretical basis to offender profiling is discussed.
Article
Full-text available
University students, police professionals, and a logistic regression model were provided with information on 38 pairs of burglaries, 20% of which were committed by the same offender, in order to examine their ability to accurately identify linked serial burglaries. For each offense pair, the information included: (1) the offense locations as points on a map, (2) the distance (in km) between the two offenses, (3) entry methods, (4) target characteristics, and (5) property stolen. Half of the participants received training informing them that the likelihood of two offenses being committed by the same offender increases as the distance between the offenses decreases. Results showed that students outperformed police professionals, that training increased decision accuracy, and that the logistic regression model achieved the highest rate of success. Potential explanations for these results are presented, focusing primarily on the participants' use of offense information, and their implications are discussed.
Article
Full-text available
The aim of the present study was to investigate whether experienced car crime investigators have special expertise in the linking of car crime and what characterises effective decision‐making in this context. Groups of experienced car crime investigators, experienced other investigators, novice participants, and naive participants attempted to link 10 series of three car crimes while thinking aloud. The results showed that experience had an effect on actual and self‐assessed linking accuracy but not on processing speed. Linking accuracy was also related to the use of a limited subset of case characteristics. Characteristics used in successful linking were included in a multidimensional scaling analysis which showed that these characteristics could be used to link the cases together also automatically. The implications of the findings for car crime investigation and for the creation of automated decision‐support systems were discussed.
Article
Full-text available
Purpose. The current study tests whether existing behavioural case linkage findings from the United Kingdom (UK) will replicate abroad with a sample of residential burglaries committed in Finland. In addition, a previously discussed methodological issue is empirically explored. Methods. Seven measures of behavioural similarity, geographical proximity, and temporal proximity are calculated for pairs of burglary crimes committed by 117 serial burglars in Finland. The ability of these seven measures to distinguish between pairs of crimes committed by the same offender (linked pairs) and different offenders (unlinked pairs) is tested using logistic regression and receiver operating characteristic (ROC) analysis. Two methodologies for forming the unlinked pairs are compared; one representing the ‘traditional’ approach used by research and, the other, a new approach that represents a potentially more realistic and statistically sound approach to testing case linkage. Results. A wider range of offender behaviours were able to distinguish between linked and unlinked crime pairs in the current Finnish sample than in previous UK-based research. The most successful features were the kilometre-distance between crimes (the intercrime distance), the number of days separating offences (temporal proximity), and a combination of target, entry, internal, and property behaviours (the combined domain). There were no statistically significant differences between the two methodological approaches. Conclusions. The current findings demonstrate that a wider range of offender behaviours can be used to discriminate between linked and unlinked residential burglary crimes committed in Finland than in the UK. The use of a more realistic and statistically sound methodology does not lead to substantial changes in case linkage findings.
Article
Full-text available
Purpose. This paper is concerned with case linkage, a form of behavioural analysis used to identify crimes committed by the same offender, through their behavioural similarity. Whilst widely practised, relatively little has been published on the process of linking crimes. This review aims to draw together diverse published studies by outlining what the process involves, critically examining its underlying psychological assumptions and reviewing the empirical research conducted on its viability.Methods. Literature searches were completed on the electronic databases, PsychInfo and Criminal Justice Abstracts, to identify theoretical and empirical papers relating to the practice of linking crimes and to behavioural consistency.Results. The available research gives some support to the assumption of consistency in criminals' behaviour. It also suggests that in comparison with intra-individual variation in behaviour, inter-individual variation is sufficient for the offences of one offender to be distinguished from those of other offenders. Thus, the two fundamental assumptions underlying the practice of linking crimes, behavioural consistency and inter-individual variation, are supported. However, not all behaviours show the same degree of consistency, with behaviours that are less situation-dependent, and hence more offender-initiated, showing greater consistency.Conclusions. The limited research regarding linking offenders' crimes appears promising at both a theoretical and an empirical level. There is a clear need, however, for replication studies and for research with various types of crime.
Article
Full-text available
The purpose of this study is to determine if readily available information about commercial and residential serial burglaries, in the form of the offender's modus operandi, provides a statistically significant basis for accurately linking crimes committed by the same offender. Logistic regression analysis is applied to examine the degree to which various linking features can be used to discriminate between linked and unlinked burglaries. Receiver operating characteristic (ROC) analysis is then performed to calibrate the validity of these features and to identify optimal decision thresholds for linking purposes. Contrary to crime scene behaviours traditionally examined to link serial burglaries, the distance between crime site locations demonstrated significantly greater effectiveness as a linking feature for both commercial and residential burglaries. Specifically, shorter distances between crimes signalled an increased likelihood that burglaries were linked. Thus, these results indicate that, if one examines suitable behavioural domains, high levels of stability and distinctiveness exist in the actions of serial burglars, and these actions can be used to accurately link crimes committed by the same offender. Copyright © 2005 John Wiley & Sons, Ltd.
Article
Full-text available
Purpose. Through an examination of serial rape data, the current article presents arguments supporting the use of receiver operating characteristic (ROC) analysis over traditional methods in addressing challenges that arise when attempting to link serial crimes. Primarily, these arguments centre on the fact that traditional linking methods do not take into account how linking accuracy will vary as a function of the threshold used for determining when two crimes are similar enough to be considered linked.Methods. Considered for analysis were 27 crime scene behaviours exhibited in 126 rapes, which were committed by 42 perpetrators. Similarity scores were derived for every possible crime pair in the sample. These measures of similarity were then subjected to ROC analysis in order to (1) determine threshold-independent measures of linking accuracy and (2) set appropriate decision thresholds for linking purposes.Results. By providing a measure of linking accuracy that is not biased by threshold placement, the analysis confirmed that it is possible to link crimes at a level that significantly exceeds chance (AUC = .75). The use of ROC analysis also allowed for the identification of decision thresholds that resulted in the desired balance between various linking outcomes (e.g. hits and false alarms).Conclusions. ROC analysis is exclusive in its ability to circumvent the limitations of threshold-specific results yielded from traditional approaches to linkage analysis. Moreover, results of the current analysis provide a basis for challenging common assumptions underlying the linking task.
Article
Full-text available
Since the 1970s, a wide body of research has suggested that the accuracy of clinical risk assessments of violence might be increased if clinicians used actuarial tools. Despite considerable progress in recent years in the development of such tools for violence risk assessment, they remain primarily research instruments, largely ignored in daily clinical practice. We argue that because most existing actuarial tools are based on a main effects regression approach, they do not adequately reflect the contingent nature of the clinical assessment processes. To enhance the use of actuarial violence risk assessment tools, we propose a classification tree rather than a main effects regression approach. In addition, we suggest that by employing two decision thresholds for identifying high- and low-risk cases--instead of the standard single threshold--the use of actuarial tools to make dichotomous risk classification decisions may be further enhanced. These claims are supported with empirical data from the MacArthur Violence Risk Assessment Study.
Article
Full-text available
Previous studies that have compared logistic regression (LR), classification and regression tree (CART), and neural networks (NNs) models for their predictive validity have shown inconsistent results in demonstrating superiority of any one model. The three models were tested in a prospective sample of 1225 UK male prisoners followed up for a mean of 3.31years after release. Items in a widely-used risk assessment instrument (the Historical, Clinical, Risk Management-20, or HCR-20) were used as predictors and violent reconvictions as outcome. Multi-validation procedure was used to reduce sampling error in reporting the predictive accuracy. The low base rate was controlled by using different measures in the three models to minimize prediction error and achieve a more balanced classification. Overall accuracy of the three models varied between 0.59 and 0.67, with an overall AUC range of 0.65–0.72. Although the performance of NNs was slightly better than that of LR and CART models, it did not demonstrate a significant improvement. KeywordsViolence reconviction–Risk assessment–Neural networks–Classification and regression tree–HCR-20
Article
Full-text available
Case linkage involves identifying crime series on the basis of behavioral similarity and distinctiveness. Research regarding the behavioral consistency of serial rapists has accumulated; however, it has its limitations. One of these limitations is that convicted or solved crime series are exclusively sampled whereas, in practice, case linkage is applied to unsolved crimes. Further, concerns have been raised that previous studies might have reported inflated estimates of case linkage effectiveness due to sampling series that were first identified based on similar modus operandi (MO), thereby overestimating the degree of consistency and distinctiveness that would exist in naturalistic settings. We present the first study to overcome these limitations; we tested the assumptions of case linkage with a sample containing 1) offenses that remain unsolved, and 2) crime series that were first identified as possible series through DNA matches, rather than similar MO. Twenty-two series consisting of 119 rapes from South Africa were used to create a dataset of 7021 crime pairs. Comparisons of crime pairs that were linked using MO vs. DNA revealed significant, but small differences in behavioral similarity with MO-linked crimes being characterized by greater similarity. When combining these two types of crimes together, linked pairs (those committed by the same serial offender) were significantly more similar in MO behavior than unlinked pairs (those committed by two different offenders) and could be differentiated from them. These findings support the underlying assumptions of case linkage. Additional factors thought to impact on linkage accuracy were also investigated. KeywordsComparative case analysis–Linkage analysis–Behavioral linking–Sexual assault–Sexual offense
Article
Full-text available
This article examines a process of behavioral analysis, referred to as linkage analysis, used in identifying sexual offenses that have been committed by the same offender. This type of analysis examines behavior that is contained in three distinct components of a crime, i.e., the modus operandi (MO) or the “how to” of a crime; the ritual or fantasy-based behaviors for a particular type or series of sexual crimes; the signature or unique combination of behaviors, which suggests that a series of crimes has been perpetrated by the same offender. Linkage analysis involves five assessment procedures: (1) gathering detailed, varied, and multisource documentation; (2) reviewing the documentation and identifying significant features of each crime individually across the series; (3) classifying the significant features of the crime as either MO and/or ritualistic constructs; (4) comparing the combination of MO and ritualistic features across the crimes to determine if a signature exists; (5) compiling a written analysis that details the conclusions derived from the available information. Results of this type of analysis can be used for investigative purposes and, in some instances, can help to inform the decision-making of the courts.
Article
Full-text available
Citation: Monahan, J., Steadman, H., Silver, E., Appelbaum, P., Robbins, P., Mulvey, E., Roth, L., Grisso, T., & Banks, S. (2001). Rethinking risk assessment: The MacArthur study of mental disorder and violence. New York: Oxford University Press. ISBN 0195138821, 9780195138825. Winner of the American Psychiatric Association's Manfred S. Guttmacher Award, 2002. Publisher summary: The presumed link between mental disorder and violence has been the driving force behind mental health law and policy for centuries. Legislatures, courts, and the public have come to expect that mental health professionals will protect them from violent acts by persons with mental disorders. Yet for three decades research has shown that clinicians' unaided assessments of "dangerousness" are barely better than chance. Rethinking Risk Assessment: The MacArthur Study of Mental Disorder and Violence tells the story of a pioneering investigation that challenges preconceptions about the frequency and nature of violence among persons with mental disorders, and suggests an innovative approach to predicting its occurrence. The authors of this massive project -- the largest ever undertaken on the topic -- demonstrate how clinicians can use a "decision tree" to identify groups of patients at very low and very high risk for violence. This dramatic new finding, and its implications for the every day clinical practice of risk assessment and risk management, is thoroughly described in this remarkable and long-anticipated volume. Taken to heart, its message will change the way clinicians, judges, and others who must deal with persons who are mentally ill and may be violent will do their work. Preview available via Google Books.
Article
Full-text available
Thesis submitted for the degree of Doctor of Philosophy at the University of Leicester, December 2008. Awarded 23 October 2009. The first chapter of the thesis critically reviews the research on juvenile violent and sexual offending and highlights the heterogeneity of such offenders in terms of those that persist and those that assault different types of victim. Research on juvenile stranger sex offenders and their offence characteristics is explored. Chapter 2 presents empirical research on the behavioural consistency and distinctiveness of juvenile stranger sex offending and whether case linkage can accurately identify the crimes of serial offenders. Calls from personality psychologists to consider the context of behaviour when investigating behavioural consistency are responded to with preliminary research into incorporating context in case linkage. Evidence for behavioural consistency and distinctiveness is reported for serial juvenile stranger sex offenders, however evidence for consistency in ‘if(victim behaviour)-then(offender behaviour)’ contingencies is less convincing. Chapter 3 investigates ways of prioritising sex offences for crime analysis. Whether juvenile serial stranger sex offenders escalate in their use of physical aggression is investigated with few “increasers” being identified. Preliminary findings suggest some characteristics on which increasers vs. non-increasers differ that might inform investigative risk assessment. However, escalation appears largely related to learning behaviour and progression to more elaborate sexual assaults. Preliminary findings suggest some offence behaviours that appear more characteristic of offences occurring later in a series. Chapter 4 investigates and contrasts group rape by juvenile and adult perpetrators. How applicable social psychological theories of group violence are to group rape is tested with findings suggesting that theories of group dynamics as well as social convergence are relevant. Further, aggression in group rapes appears both expressive and instrumental in purpose. Roles adopted by group members are investigated. Evidence of distinct leaders and followers in group rapes is identified using both Porter and Alison’s (2001) Scale of Influence and through the use of pragmatics theory. Additional roles are discussed.
Article
Full-text available
The present study investigated the possibility of statistically linking arson cases based on consistency of behaviors from one crime scene to another. Serial and spree arson cases were studied to differentiate underlying themes and to link cases committed by the same offender. The material consisted of 248 arson cases which formed 42 series of arsons. A content analysis using 45 dichotomous variables was carried out and principal components (PCA) analysis was performed to identify underlying themes. Summary scores reflecting the themes were calculated. Linking effectiveness was tested with a discriminant analysis using the summary scores. The PCA analysis was successful and underlying themes which were in accordance with previous studies could be identified. Six factors were retained, in the PCA. The linking of the arson cases was possible to a satisfactory level: 33% of the cases could be correctly linked and for over 50% of the cases, the series they actually belonged to was among the ten series identified as most probable on the basis of the linking analysis. From a practical point of view, the results could be used as a basis for developing support systems for police investigations of arson. © Copyright Springer, 2004
Article
Full-text available
We performed a Monte Carlo study to evaluate the effect of the number of events per variable (EPV) analyzed in logistic regression analysis. The simulations were based on data from a cardiac trial of 673 patients in which 252 deaths occurred and seven variables were cogent predictors of mortality; the number of events per predictive variable was (252/7 =) 36 for the full sample. For the simulations, at values of EPV = 2, 5, 10, 15, 20, and 25, we randomly generated 500 samples of the 673 patients, chosen with replacement, according to a logistic model derived from the full sample. Simulation results for the regression coefficients for each variable in each group of 500 samples were compared for bias, precision, and significance testing against the results of the model fitted to the original sample. For EPV values of 10 or greater, no major problems occurred. For EPV values less than 10, however, the regression coefficients were biased in both positive and negative directions; the large sample variance estimates from the logistic model both overestimated and underestimated the sample variance of the regression coefficients; the 90% confidence limits about the estimated values did not have proper coverage; the Wald statistic was conservative under the null hypothesis; and paradoxical associations (significance in the wrong direction) were increased. Although other factors (such as the total number of events, or sample size) may influence the validity of the logistic model, our findings indicate that low EPV can lead to major problems.
Article
Full-text available
This paper uses statistical models to test directly the police practice of utilising modus operandi to link crimes to a common offender. Data from 86 solved commercial burglaries committed by 43 offenders are analysed using logistic regression analysis to identify behavioural features that reliably distinguish between linked and unlinked crime pairs. Receiver operating characteristic analysis is then used to assign each behavioural feature an overall level of predictive accuracy. The results indicate that certain features, in particular the distances between burglary locations, lead to high levels of predictive accuracy. This study therefore reveals some of the important consistencies in commercial burglary behaviour. These have theoretical value in helping to explain criminal activity. They also have practical value by providing the basis for a diagnostic tool that could be used in comparative case analysis.
Article
Full-text available
Classification trees based on exhaustive search algorithms tend to be biased towards selecting variables that afford more splits. As a result, such trees should be interpreted with caution. This article presents an algorithm called QUEST that has negligible bias. Its split selection strategy shares similarities with the FACT method, but it yields binary splits and the final tree can be selected by a direct stopping rule or by pruning. Real and simulated data are used to compare QUEST with the exhaustive search approach. QUEST is shown to be substantially faster and the size and classification accuracy of its trees are typically comparable to those of exhaustive search. 1
Article
Classification trees based on exhaustive search algorithms tend to be biased towards selecting variables that afford more splits. As a result, such trees should be interpreted with caution. This article presents an algorithm called QUEST that has negligible bias. Its split selection strategy shares similarities with the FACT method, but it yields binary splits and the final tree can be selected by a direct stopping rule or by pruning. Real and simulated data are used to compare QUEST with the exhaustive search approach. QUEST is shown to be substantially faster and the size and classification accuracy of its trees are typically comparable to those of exhaustive search.
Article
Analysis and interpretation of higher order cross-tabulation data are of recurring concern in marketing research. The authors present a parsimonious new approach to this data analysis problem. Specifically, a model-free approach is proposed which helps to identify differences in the response distribution of a criterion variable based on segments of respondents defined by characteristics of the predictor variables. The approach, which relies on Bonferroni adjusted chi square statistics to direct a sequential search process, is illustrated in a segmentation analysis of data from a national consumer survey. The results of analyzing the same data by using AID and LOGIT procedures are also examined. The article concludes with a discussion of potential applications, limitations, and extensions of the new approach.
Article
In the absence of physical evidence, investigators must often rely on offence behaviours when determining whether several crimes are linked to a common offender. A variety of factors can potentially influence the degree to which accurate linking is possible, including the similarity coefficient used to assess across-crime similarity. The current study examines the performance of two similarity coefficients that have recently been compared to one another, Jaccard's coefficient (J) and the taxonomic similarity index (Δs), using samples of two crime types, serial homicide (N=237) and serial burglary (N=210). In contrast to previous research, the results indicate that Δs does not significantly outperform J with respect to linking accuracy. In addition, both coefficients lead to higher levels of linking accuracy in cases of serial homicide compared to serial burglary. Potential explanations for these findings are presented and their implications are discussed.
Article
Evidence about a suspect's behavioural similarity across a series of crimes has been presented in legal proceedings in at least three different countries. Its admission as expert evidence, whilst still rare, is becoming more common thus it is important for us to understand how such evidence is received by jurors and legal professionals. This article reports on a qualitative analysis of mock jurors' deliberations about expert linkage analysis evidence. Three groups of mock jurors (N = 20) were presented with the prosecution's linkage analysis evidence from the USA State v. Fortin I murder trial and expert evidence for the defence constructed for the purposes of the study. Each group was asked to deliberate and reach a verdict. Deliberations were video-recorded and subject to thematic content analysis. The themes that emerged were varied. Analysis suggested that the mock jurors were cautious of the expert evidence of behavioural similarity. In some cases they were sceptical of the expert. They articulated a preference that expert opinion be supported using statistics. Additional themes included jurors having misconceptions concerning what is typical offender behaviour during rape which suggests there is a need for expert linkage analysis evidence regarding behavioural similarities and the relative frequencies of crime scene behaviours. Copyright © 2010 John Wiley & Sons, Ltd.
Article
A sample of serial stranger rape cases ( n = 43) that had occurred in Finland during the years 1983–2001 were studied with the objectives being to: (a) describe the characteristics of the offenders; (b) explore the structure of serial rape; and (c) demonstrate behavioural linkage through an analysis of the offenders' crime scene behaviour using both multidimensional scaling (MDS) and discriminant function analysis (DFA). The material was content analysed with regard to the occurrence of a number of dichotomous variables. The inter-relationships of the variables was studied using MDS. The analysis revealed two previously identified major modes of interaction with the victim: involvement and hostility. Employing MDS and DFA, it was shown that the offences of different offenders were distinguishable in terms of variation between the offences of different offenders and consistency within the offences of a single offender. Using DFA, the classification accuracy clearly exceeds that expected by chance, and 25.6% of the cases were classified without any error. The results are discussed in relation to their practical utility and previous studies. Copyright © 2005 John Wiley & Sons, Ltd.
Article
Whilst case linkage is used with serious forms of serial crime (e.g. rape and murder), the potential exists for it to be used with volume crime. This study replicates and extends previous research on the behavioural linking of burglaries. One hundred and sixty solved residential burglaries were sampled from a British police force. From these, 80 linked crime pairs (committed by the same serial offender) and 80 unlinked crime pairs (committed by two different serial offenders) were created. Following the methodology used by previous researchers, the behavioural similarity, geographical proximity, and temporal proximity of linked crime pairs were compared with those of unlinked crime pairs. Geographical and temporal proximity possessed a high degree of predictive accuracy in distinguishing linked from unlinked pairs as assessed by logistic regression and receiver operating characteristic analyses. Comparatively, other traditional modus operandi behaviours showed less potential for linkage. Whilst personality psychology literature has suggested we might expect to find a relationship between temporal proximity and behavioural consistency, such a relationship was not observed. Copyright © 2010 John Wiley & Sons, Ltd.
Article
Receiver operating characteristic (ROC) analysis is a widely used and accepted method for improving decision making performance across a range of diagnostic settings. The goal of this paper is to demonstrate how ROC analysis can be used to improve the quality of decisions made routinely in a policing context. To begin, I discuss the general principles underlying the ROC approach and demonstrate how one can conduct the analysis. Several practical applications of ROC analysis are then presented by drawing on a number of policing tasks where the procedure has been used already (bite mark identification and linking serial crimes) or where it could be used in the future (statement validity assessment and determining the veracity of suicide notes). I conclude by considering briefly some of the potential difficulties that may be encountered when using ROC analysis in the policing context and offer some possible solutions to these problems. Copyright © 2005 John Wiley & Sons, Ltd.
Article
This is a progress report on the development of practical methods for the actuarial prediction of violence. The literature indicates that actuarial prediction is more accurate than clinical prediction, but in practice actuarial methods seem to be used rarely. Here we address two obstacles to the clinical use of actuarial prediction methods. First, clinicians may be averse to actuarial methods that require calculations. To remedy this, we developed a regression tree screen that presents actuarial information about violence in a series of yes/no questions. Second, using actuarial methods to identify the small minority of violent patients in a general psychiatric population may be too costly. To remedy this, we developed a method to prescreen patients for intensive evaluation using an inexpensive assessment. We evaluated regression trees and two-stage screening by comparing their accuracies against conventional actuarial methods. The results showed that actuarial predictions based on regression trees and two-stage screens were as accurate as regression-based methods in identifying repetitively violent patients. These easier-to-use methods may therefore be useful techniques for actuarial predictions.
Article
This is an account of what I have learned (so far) about the application of statistics to psychology and the other sociobiomedical sciences. It includes the principles "less is more" (fewer variables, more highly targeted issues, sharp rounding off), "simple is better" (graphic representation, unit weighting for linear com- posites), and "some things you learn aren't so." I have learned to avoid the many misconceptions that surround Fisherian null hypothesis testing. I have also learned the importance of power analysis and the determination of just how big (rather than how statistically significant) are the effects that we study. Finally, I have learned that there is no royal road to statistical induction, that the informed judgment of the investigator is the crucial element in the interpretation of data, and that things take time.
Article
The purpose of the present study is to test the case linkage principles of behavioural consistency and behavioural distinctiveness using serial vehicle theft data. Data from 386 solved vehicle thefts committed by 193 offenders were analysed using Jaccard's, regression and Receiver Operating Characteristic analyses to determine whether objectively observable aspects of crime scene behaviour could be used to distinguish crimes committed by the same offender from those committed by different offenders. The findings indicate that spatial behaviour, specifically the distance between theft locations and between dump locations, is a highly consistent and distinctive aspect of vehicle theft behaviour; thus, intercrime and interdump distance represent the most useful aspects of vehicle theft for the purpose of case linkage analysis. The findings have theoretical and practical implications for understanding of criminal behaviour and for the development of decision-support tools to assist police investigation and apprehension of serial vehicle theft offenders.
Article
Jaccard has been the choice similarity metric in ecology and forensic psychology for comparison of sites or offences, by species or behaviour. This paper applies a more powerful hierarchical measure - taxonomic similarity (s), recently developed in marine ecology - to the task of behaviourally linking serial crime. Forensic case linkage attempts to identify behaviourally similar offences committed by the same unknown perpetrator (called linked offences). s considers progressively higher-level taxa, such that two sites show some similarity even without shared species. We apply this index by analysing 55 specific offence behaviours classified hierarchically. The behaviours are taken from 16 sexual offences by seven juveniles where each offender committed two or more offences. We demonstrate that both Jaccard and s show linked offences to be significantly more similar than unlinked offences. With up to 20% of the specific behaviours removed in simulations, s is equally or more effective at distinguishing linked offences than where Jaccard uses a full data set. Moreover, s retains significant difference between linked and unlinked pairs, with up to 50% of the specific behaviours removed. As police decision-making often depends upon incomplete data, s has clear advantages and its application may extend to other crime types. Copyright © 2007 John Wiley & Sons, Ltd.
Article
Case linkage, the identification of crimes suspected of being committed by the same perpetrator on the basis of behavioral similarity, and offender profiling, the inference of offender characteristics from offense behaviors, are used to advise police investigations and, in relation to case linkage, have been admitted in legal proceedings. Criteria for expert evidence, such as the Daubert criteria (Daubert v. Merrell Dow Pharmaceuticals, 1993), place stringent conditions on the admissibility of expert evidence. The future contribution of these practices to legal proceedings depends, in part, on whether they are underpinned by hypotheses that are testable and supported. The 3 hypotheses of offender behavioral consistency, of offender behavioral distinctiveness, and of a homology (direct relationship) between offender characteristics and behavior were empirically examined using a sample of serial commercial robberies. Support was found for the former 2 hypotheses but not for the last. The findings of the 2 studies have implications for the future development of these practices, for legal practitioners evaluating expert evidence, and for the implementation of public policy.
Article
Diagnostic systems of several kinds are used to distinguish between two classes of events, essentially "signals" and "noise". For them, analysis in terms of the "relative operating characteristic" of signal detection theory provides a precise and valid measure of diagnostic accuracy. It is the only measure available that is uninfluenced by decision biases and prior probabilities, and it places the performances of diverse systems on a common, easily interpreted scale. Representative values of this measure are reported here for systems in medical imaging, materials testing, weather forecasting, information retrieval, polygraph lie detection, and aptitude testing. Though the measure itself is sound, the values obtained from tests of diagnostic systems often require qualification because the test data on which they are based are of unsure quality. A common set of problems in testing is faced in all fields. How well these problems are handled, or can be handled in a given field, determines the degree of confidence that can be placed in a measured value of accuracy. Some fields fare much better than others.
Article
Advances in the field of risk assessment have highlighted the importance of developing and validating models for problematic or unique subgroups of individuals. Stalking offenders represent one such subgroup, where fears of and potential for violence are well-known and have important implications for safety management. The present study applies a Classification and Regression Tree (CART) approach to a sample of stalking offenders in order to help further the process of identifying and understanding risk assessment strategies. Data from 204 stalking offenders referred for psychiatric evaluation to a publicly-funded clinic were used to develop and assess putative risk factors. A series of nested models were used to generate tree algorithms predicting violence in this sample of offenders. Both simplified and more extensive models generated high levels of predictive accuracy that were roughly comparable to logistic regression models but much more straightforward to apply in clinical practice. Jack-knifed cross-validation analyses demonstrated considerable shrinkage in the CART, although the models were still comparable to many other actuarial risk assessment instruments. Logistic regression models were much more resilient to cross-validation, with relatively modest loss in predictive power.
Article
The application of statistical modeling techniques, including classification and regression trees, in the prediction of violence has increasingly received attention. The predictive performance of logistic regression and classification tree methods in predicting violence was explored in a sample of patients with psychotic illness. Of 2 logistic regression models, the forward stepwise method produced a simpler model than the full model, but the latter performed better. The performance of the classification tree appeared to be high before cross-validation, but reduced when cross-validated. The standard logistic model was the most robust model. A simplified tree with extra weight given to violent cases was a reasonable competitor and was simple to apply. Although classification trees can be suitable for routine clinical practice, because of the simplicity of their decision-making processes, their robustness and therefore clinical utility was problematic in this sample. Further research is required to compare such models in large prospective epidemiologic studies of other psychiatric populations.
Article
Clinicians and health service researchers are frequently interested in predicting patient-specific probabilities of adverse events (e.g. death, disease recurrence, post-operative complications, hospital readmission). There is an increasing interest in the use of classification and regression trees (CART) for predicting outcomes in clinical studies. We compared the predictive accuracy of logistic regression with that of regression trees for predicting mortality after hospitalization with an acute myocardial infarction (AMI). We also examined the predictive ability of two other types of data-driven models: generalized additive models (GAMs) and multivariate adaptive regression splines (MARS). We used data on 9484 patients admitted to hospital with an AMI in Ontario. We used repeated split-sample validation: the data were randomly divided into derivation and validation samples. Predictive models were estimated using the derivation sample and the predictive accuracy of the resultant model was assessed using the area under the receiver operating characteristic (ROC) curve in the validation sample. This process was repeated 1000 times-the initial data set was randomly divided into derivation and validation samples 1000 times, and the predictive accuracy of each method was assessed each time. The mean ROC curve area for the regression tree models in the 1000 derivation samples was 0.762, while the mean ROC curve area of a simple logistic regression model was 0.845. The mean ROC curve areas for the other methods ranged from a low of 0.831 to a high of 0.851. Our study shows that regression trees do not perform as well as logistic regression for predicting mortality following AMI. However, the logistic regression model had performance comparable to that of more flexible, data-driven models such as GAMs and MARS.
Article
Unlabelled: Residential location of a criminal can be predicted statistically [M. Laukkanen, P. Santtila, Predicting the home location of a serial commercial robber, Forensic Sci. Int. 157 (2006) 71-82]. Examined were: accuracy of the technique for urban burglary series, correlations between way of committing burglary and distance and use of those correlations in enhancing prediction accuracy. Data: 78 residential burglary series from Greater Helsinki area, Finland. Series for which the home location prediction was made was never part of the predicting model. Distances between home and crime-site were short (Mdn 3.88km; IQR=1.16-10.10km). Search area of a perpetrator could be limited to 1.95% (Mdn, IQR=0.64-18.70%) of the total study area. For series which conformed to Circle Hypothesis (45%), search area was 0.84% (Mdn, IQR=0.51-2.34%). Correlations between crime features and distance were found to enhance accuracy when features of series hinted short distance: sub-model limited search area to 0.19% (Mdn, IQR=0.07-0.65%).
Investigating individual differences in the expression of behavioural consistency in crime series using ICT analyses
  • C Bennell
  • J Woodhams
  • E Beauregard
Bennell, C., Woodhams, J., & Beauregard, E. (in preparation). Investigating individual differences in the expression of behavioural consistency in crime series using ICT analyses.
The use of a linkage analysis as an investigative tool and evidential material in serial offenses
  • G Labuschagne
Labuschagne, G. (2012). The use of a linkage analysis as an investigative tool and evidential material in serial offenses. In K. Borgeson, & K. Kuehnle (Eds.), Serial offenders: Theory and practice (pp. 187-215). Sudbury, MA: Jones & Bartlett Learning.
The linking of burglary crimes using offender behavior: Testing research cross-nationally and in more realistic settings. Legal and Criminological Psychology Advance online publication Juvenile sex offending: An investigative perspective (Unpublished doctoral dissertation)
  • M Tonkin
  • P Santtila
  • R Bull
Tonkin, M., Santtila, P., & Bull, R. (2011). The linking of burglary crimes using offender behavior: Testing research cross-nationally and in more realistic settings. Legal and Criminological Psychology. Advance online publication. DOI: 10.1111/j.2044-8333.2010.02007.x Woodhams, J. (2008). Juvenile sex offending: An investigative perspective (Unpublished doctoral dissertation). University of Leicester, Leicester, UK.
The psychology of linking crimes: A review of the evidence. Legal and Criminological Psychology A test of case linkage principles with solved and unsolved serial rapes
  • J Woodhams
  • C R Hollin
  • R J Bull
  • G Labuschagne
Woodhams, J., Hollin, C. R., & Bull, R. (2007). The psychology of linking crimes: A review of the evidence. Legal and Criminological Psychology, 12, 233–249. DOI: 10.1348/135532506X118631 Woodhams, J., & Labuschagne, G. (2012). A test of case linkage principles with solved and unsolved serial rapes. Journal of Police and Criminal Psychology, 27, 85–98. DOI: 10.1007/s11896-011- 9091-1
  • M Tonkin
M. Tonkin et al. Copyright © 2012 John Wiley & Sons, Ltd. J. Investig. Psych. Offender Profil. 9: 235–258 (2012) DOI: 10.1002/jip
Linking serious sexual assaults through behaviour (Home Office Research Study 215)
  • D Grubin
  • P Kelly
  • C Brunsdon
Grubin, D., Kelly, P., & Brunsdon, C. (2001). Linking serious sexual assaults through behaviour (Home Office Research Study 215). London, UK: Home Office Research, Development and Statistics Directorate.
Are all serial rapists consistent in the same way
  • J Woodhams
  • C Bennell
  • E Beauregard
Woodhams, J., Bennell, C., & Beauregard, E. (2011). Are all serial rapists consistent in the same way? In M. Tonkin (Chair), Linking crimes using offender behaviour: New and emerging directions for research. Symposium conducted at the 20 th Annual Division of Forensic Psychology Conference 2011, Portsmouth, UK.