Edwards, Michael J.2021-04-152021-04-152020https://hdl.handle.net/10092/101802http://dx.doi.org/10.26021/10856The main aim of this thesis was to contribute to the understanding of arson offending in terms of arson risk factors, arson recidivism, arson actuarial risk assessment and whether arson recidivists are qualitatively different from other types of recidivists (such as violent and non-violent offenders). The findings from this thesis determined that second-generation actuarial tools can be successfully developed to predict rare offending events such as arson recidivism and more importantly become operationally viable tools to assist multiple sectors in the criminal justice system such as judicial, treatment, custodial, parole and investigations. Prior to our published work contained within this thesis (see chapter 2) there were no empirically developed, validated or publishable work in New Zealand (NZ) or internationally on second-generation actuarial tools for arson recidivism. This led to the overarching aim of this thesis which was to develop arson predictive models and arson actuarial tools to aid the prediction of arson recidivism in a New Zealand context. The research presented in this thesis sets the benchmark for researchers to replicate and develop future arson predictive models and operationalised actuarial risk assessment tools for arson recidivism within their respective jurisdictions. Chapter one provides a literature review with the aim of presenting key arson research findings and how this background knowledge supports the overarching goal and aims of this thesis. In this chapter we define arson and its problem in the US, UK and NZ. We discuss the four generations of risk assessments as defined by Bonta (1996). In this sub-topic we discuss the Risk of re-Conviction and Risk of re-Imprisonment (RoC*RoI; Bakker, Riley, & O’Malley, 1999) as a preferred second-generation actuarial measure of choice for NZ offenders in a custodial setting. We progress to third generation approaches and review several promising firesetting risk assessment tools. We complete this topic by discussing the Risk Need Responsivity Model (RNR; Bonta & Andrews, 2007) and how it guides second, third and fourth-generation risk assessment approaches. From here we discuss several firesetting theories; the Dynamic Behavioural Theory of Firesetting (DBToF; Fineman, 1980), Kolko and Kazdin (1986) social learning model and the Multi-Trajectory Theory of Adult Firesetting (M-TTAF; Gannon, Ó Ciardha, Doley, & Alleyne, 2012). We briefly touch on other recently developed micro-theories and UK firesetting intervention programmes (FIPP; FIP-MO) that are guided by the M-TTAF theory. Next, we discuss arson recidivism and arson risk factors from fifteen published arson studies (between 1978 and 2018) of which have guided and informed our research. From here we reviewed and compared the work of four key published researchers who have developed arson predictive models for arson recidivism (Rice & Harris, 1996; Edwards & Grace, 2014; Field, 2015; Ducat, McEwan, & Ogloff, 2015). Two of these publishers (Edwards & Grace, 2014; Field, 2015) have progressed and developed operationalised actuarial tools for arson recidivism. Last, we briefly discussed an arson classification table based on the style and type of offending (the serial, mass or spree arsonist). Following this overview and background within the field, we progressed to chapter two which comprised of the original work by Edwards and Grace (2014). Given that there was no published research in the literature relating to second-generation arson actuarial development the rationale for conducting this original piece of research was certainly warranted and worthy of exploration. As such, chapter two was founded on this rationale which forms the complete chapter titled “The Development of an Actuarial Model for Arson Recidivism” (Edwards & Grace, 2014). Our work was based on previous research conducted by Rice and Harris (1996) who investigated mentally disordered arsonists. Using their research design as a framework we developed empirical-based predictive models and actuarial tools for arson recidivism among NZ convicted arson offenders. Our research studied individuals who were prosecuted through the NZ criminal justice system for an arson-related offence in NZ between 1985 and 1994 (n = 1250). Over a 10-year follow-up, recidivism rates for arson were 6.2%, violent 48.5% and non-violent 79.3%. A major goal of this study was to develop predictive models for arson, violent and non-violent recidivism. The final predictors for the arson model were: First arson under 18-years, multiple arsons and having prior vandalism offences. In comparison, the final predictors for the violent model were: First arson under 18- years, age at first offence, number of prior violent and prior all offences; and for the non- violent model: Age at first arson, number of prior theft and number of prior drug offences. Overall, these findings suggest that arson recidivists have specific-risk predictors that are not routinely found in violent and non-violent recidivists. It is concluded that arson recidivists are qualitatively different from offenders with non-arson criminal histories. The empirical evidence presented suggests that arson recidivists should be cautiously considered as a distinct and unique category of re-offending. For this reason, it is important to examine specific risk predictors that have been empirically validated to predict future arson offending. In terms of the model accuracy, the arson model is operating at a moderate level (AUC = .68) compared to the violent and non-violent models which are operating at slightly higher levels of predictive accuracy (AUC = .72 and .73, respectively). The final goal of the study was to develop an operationalised second-generation actuarial risk assessment tool for identifying “high-risk” individuals who are significantly more likely to commit an arson offence in the future. The actuarial tool was based on the same three final risk predictors that generated the arson model with the defined risk scale varying from low to high (0 to 10). The arson actuarial tool provided a moderate level of predictive accuracy (AUC = .67). These results hold great promise for clinicians and practitioners to incorporate the Edwards and Grace (2014) arson actuarial tool as part of their comprehensive risk assessment and case management plans for third and fourth-generation approaches. It is interesting to note that since the published work by Edwards and Grace (2014) other researchers such as Field (2015) have replicated and developed additional arson predictive models and arson actuarial tools using similar methodology by Edwards and Grace (2014). In chapter three, we replicated the original work by Edwards and Grace (2014) and developed additional empirical-based predictive models for arson, violent and non-violent recidivism and a subsequent arson actuarial model. The rationale for this second study was to assess the generalisability and utility of the original Edwards and Grace (2014) model against a second arson cohort series (with no overlapping dates). To achieve this, we obtained and investigated a second NZ sample of arson offenders who were convicted of an arson-related offence in NZ between 1998 and 2008 (n = 1464) and a random sample of convicted violent (n = 1464) and non-violent offenders (n = 1464). Over a 5-year follow-up, recidivism rates for arson were 5.9%, for violence (violent sample) 51% and for non-violence (non-violent sample) 72.5%. Similarly, we identified and compared the final static risk predictors associated with arson, violent and non-violent recidivism. We developed predictive models for each recidivism type and identified the final predictors for the arson model were: First arson under 18-years and prior arson offences. We replicated and built a comparative second-generation actuarial tool for arson recidivism using the same three risk predictors used in the Edwards and Grace (2014) actuarial tool. Overall, both the arson predictive model and the arson actuarial tool (in chapter 3) provided low levels of accuracy (AUC = .61 and AUC = .60, respectively). Nonetheless, the two final risk predictors for the arson predictive model are reasonably well supported risk factors (Field, 2015). Therefore, the utility of the arson predictive model and its risk factors is not in serious doubt and does provide support for the development of actuarial tools for arson recidivism. It is emphasised that fine tuning research designs, methodology, using prospective data and incorporating criminogenic and dynamic risk factors is highly recommended for greater predictive accuracy and enhanced actuarial development. Last, we determined from a Linear Discriminant Analysis (LDA) test that we cannot accurately classify or distinguish a group of arson offenders from a group of violent and non-violent offenders based solely on prior criminal histories. This suggests that arson offenders are not quantitatively different compared to violent and non-violent offenders when solely comparing prior criminal histories. This supports current research that arson offenders are more criminally versatile and are not pure arsonists (Ducat et al., 2015). Lastly, the aim of the fourth chapter was to summarise the main findings from both empirical studies, its implications, future directions and limitations. We discuss and review the four critically important research questions under investigation. This chapter summarises the overarching goal, the aims within each chapter and highlights the operational utility of the Edwards and Grace (2014) tool for the NZ criminal justice system.enAll Rights ReservedTheses / Dissertations