Current literature suggests that the Level of Service Inventory (LSI) and its derivatives (LSI-R, LS/CMI, LSI-OR) are capable of predicting violent recidivism, even though they were not initially designed for this function (Girard & Wormith, 2004; Mills & Kroner, 2006). The purpose of this study was to generate violence prediction scales, based on items or subscales from the LSI-OR, using five different statistical techniques. These analyses were completed on the full construction sample, then the males and the females separately to determine how the scales differ from each other and what, if any, benefits would accrue from utilizing a gender-specific scale.<p> A cohort of 27,027 offenders who were released from custody or entered into community supervision over a one year period was included in the study. There was an average followup time of 4.4 years. In this sample there was a general recidivism rate of 36.0% and a violent recidivism rate of 11.3%. Fifteen violence prediction scales were generated that ranged in predictive validity from r = .139 to r = .214. The scale with the highest predictive validity was the 11 item scale created from the full sample using the item linear regression technique. The scale contained items indicating that history of assault, lack of education and anger management issues were related to violent recidivism. Risk levels were developed for this new scale to classify offenders from very low to very high risk.<p> Although there was little difference in the predictive validity of the generated scales, the stepwise multiple linear regression technique was identified as the most successful method of creating a tool for predicting violent recidivism. There was no increase in predictive validity when using the scale that was developed for just the females in the sample, although fewer items were consistently generated for females than males. Therefore the full sample item linear regression scale is recommended for the prediction of violent recidivism of both male and female offenders in the jurisdiction from which the data were collected. Future research directions may replicate this study in other populations and further analyze the gender differences in violent recidivism.

Current literature suggests that the Level of Service Inventory (LSI) and its derivatives (LSI-R, LS/CMI, LSI-OR) are capable of predicting violent recidivism, even though they were not initially designed for this function (Girard & Wormith, 2004; Mills & Kroner, 2006). The purpose of this study was to generate violence prediction scales, based on items or subscales from the LSI-OR, using five different statistical techniques. These analyses were completed on the full construction sample, then the males and the females separately to determine how the scales differ from each other and what, if any, benefits would accrue from utilizing a gender-specific scale.<p> A cohort of 27,027 offenders who were released from custody or entered into community supervision over a one year period was included in the study. There was an average followup time of 4.4 years. In this sample there was a general recidivism rate of 36.0% and a violent recidivism rate of 11.3%. Fifteen violence prediction scales were generated that ranged in predictive validity from r = .139 to r = .214. The scale with the highest predictive validity was the 11 item scale created from the full sample using the item linear regression technique. The scale contained items indicating that history of assault, lack of education and anger management issues were related to violent recidivism. Risk levels were developed for this new scale to classify offenders from very low to very high risk.<p> Although there was little difference in the predictive validity of the generated scales, the stepwise multiple linear regression technique was identified as the most successful method of creating a tool for predicting violent recidivism. There was no increase in predictive validity when using the scale that was developed for just the females in the sample, although fewer items were consistently generated for females than males. Therefore the full sample item linear regression scale is recommended for the prediction of violent recidivism of both male and female offenders in the jurisdiction from which the data were collected. Future research directions may replicate this study in other populations and further analyze the gender differences in violent recidivism.

Assessing offenders' risk of future violent behavior continues to be an important yet controversial role of forensic psychologists. A key debate is the relative effectiveness of assessment methods. Specifically, actuarial methods (see Quinsey et al., 1998 for a review) have been compared and contrasted to clinical and structured clinical methods (see e.g. Hart, 1998; Webster et al., 1997). Proponents of each approach argue for its superiority, yet validity studies have made few formal comparisons. In advancing the available research, the present study examines systematically the type of forensic case (i.e., sexual violence versus nonsexual violence) and type of assessment method (i.e., actuarial, structured clinical, and unstructured clinical). As observed by Borum, Otto, and Golding (1993), forensic decision making can also be influenced by the presence of certain extraneous clinical data. To address these issues, psychologists and doctoral students attending the American Psychology Law Society conference were asked to make several ratings regarding the likelihood of future sexual and nonsexual violence based on data derived from actual defendants with known outcomes. Using a mixed factorial design, each of these assessment methods were investigated for its influence on decision-makers regarding likelihood of future violence and sexually violent predator commitments. Finally, the potentially biasing effects of victim impact statements on resultant decisions were also explored.

Violence and aggression continue to be both public health and economic concerns. The field of violence prediction has undergone a series of changes in an attempt to best assess risk including using unstructured clinical judgment, actuarial measures, and structured professional judgment. Although prediction has become more accurate with improved measures, a new generation has recently emerged with an emphasis on understanding violence, as opposed to merely predicting it, to shift the focus towards violence prevention. In addition to the creation of measures, researchers have sought to identify specific risk factors for aggression and violence including static and dynamic risk factors. Despite research demonstrating associations between neuropsychological and social-cognitive factors, violence risk measures continue to omit these variables. The current study developed a multi-faceted, theoretical model of aggression including social-cognitive, neuropsychological, personality, and psychiatric factors. A community, male sample (N = 1,192) collected through Amazon's MTurk responded to a series of self-report measures and neuropsychological tasks. Utilizing structural equation modeling (SEM), I created a model predicting aggression. Several important paths were significant including from entity theory to aggression, mediated by hostile attribution bias, schizotypy to aggression, mediated by both hostile attribution bias and disinhibition, substance use to aggression mediated by disinhibition, and psychopathy to aggression directly. This model provides a framework for future research that focuses on process factors of violence and aggression.

abstract: In the artificial intelligence literature, three forms of reasoning are commonly employed to understand agent behavior: inductive, deductive, and abductive.  More recently, data-driven approaches leveraging ideas such as machine learning, data mining, and social network analysis have gained popularity. While data-driven variants of the aforementioned forms of reasoning have been applied separately, there is little work on how data-driven approaches across all three forms relate and lend themselves to practical applications. Given an agent behavior and the percept sequence, how one can identify a specific outcome such as the likeliest explanation? To address real-world problems, it is vital to understand the different types of reasonings which can lead to better data-driven inference.   This dissertation has laid the groundwork for studying these relationships and applying them to three real-world problems. In criminal modeling, inductive and deductive reasonings are applied to early prediction of violent criminal gang members. To address this problem the features derived from the co-arrestee social network as well as geographical and temporal features are leveraged. Then, a data-driven variant of geospatial abductive inference is studied in missing person problem to locate the missing person. Finally, induction and abduction reasonings are studied for identifying pathogenic accounts of a cascade in social networks. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2019