Dynamic risk factors
In the correctional domain, DRF are conceptualized as changeable factors such as intimacy deficits or emotional dysregulation that are associated with an
increased chance that individuals will reoffend. They are derived from the statistical analysis of large data sets demonstrating that certain variables are statistically associated with higher and lower rates of reoffending (Bonta and Andrews
2017). In essence, they are inductively inferred and their theoretical conceptualization is typically fairly sparse. Part of the reason for this can be traced to their origins in measures of psychological properties hypothesized to be important in causing crime and recidivism. As psychometric constructs, each DRF is really only an aggregate score representing the statistical association between different variables, which is thought to reflect a common factor or underlying latent construct (Slaney
2017). For example, Mann et al. (
2010) identified a number of empirically supported “psychologically meaningful” (DRF) risk factors for child sexual abuse which included sexual preferences for children, emotional congruence with children, offence supportive attitudes and lack of adult intimate relationships (Mann et al.
2010). While they gave brief narrative descriptions of these DRF, there was no attempt to present coherent theoretical definitions or to depict the mechanisms that cause and constitute them. In addition to their status as risk predictors, DRF are usually regarded in the correctional research and practice domain as potential causal factors that, if effectively targeted by cognitive behavioral techniques, will reduce reoffending rates. Thus they have a dual status: potential causes of crime and dynamic predictors of reoffending.
With respect to their theoretical status, researchers have identified a number of significant conceptual weaknesses in the current formulation of DRF. In the following discussion, I will consider the possible role of the three epistemic strategies outlined earlier in avoiding the problems with theoretical literacy associated with DRF research and practice. Formulating these criticisms in terms of the six types of theoretical illiteracy outlined earlier:
Uncritical and dogmatic acceptance of existing theories
DRF are viewed as possible causes of reoffending despite little evidence that they play a causal role in initial offending or recidivism rates (Ward and Fortune
2016). In addition, treatment outcome studies investigating the degree to which changes in DRF during treatment predict reduced reoffending rates have yielded disappointing—and inconsistent—results (Cording et al.
2016; Heffernan et al.
2019). The continued insistence that DRF are potential causes despite this lack of evidence reveals a degree of dogmatism and a strong commitment to risk management models of crime. It is clear that if researchers undertake research guided by a stronger theory of scientific method such as ATOM, the problems of theoretical literacy noted above are more likely to be avoided. DRF will be viewed as indicators of problems associated with crime and this will encourage researchers to consider alternative ways of conceptualizing them, possibly utilizing strength-based as well as risk-oriented models. The use of iteration (bootstrapping) techniques should introduce a degree of epistemic humility and remind researchers that their DRF-based predictive, explanatory, and treatment models are provisional at best. Finally, theory pluralism should reduce temptations to rigidly adhere to any particular methods and models of DRF and create conceptual space for the interaction between alternative ideas.
View of science as strictly empirical in nature
DRF are really just theoretical placeholders for future modeling and they have impoverished conceptual content. In addition, they represent different types of possible harm and thus are heavily value laden (e.g., the DRF intimacy deficits is only considered a risk factor because it is associated with harm to victims, a disvalued condition). The use of ATOM will remind researchers that science has a strong conceptual dimension as well as an empirical one, and encourage them to think explicitly about the mechanisms causing and constituting DRF. Exploring DRF and non-crime related problems in a dynamic way (epistemic iteration) over time may reveal commonalties between them and result in a focus on the individuals’ core problems such as threat appraisal dysfunction, rather than artificially dividing them into anxiety symptoms and aggressive behavior. Theory pluralism extends to values and should prompt researchers (and clinicians) to think explicitly about what type of value conflicts are associated reoffending and how they impact on individuals’ behavior.
Undervaluing the importance of conceptual analysis
DRF suffer from the conceptual problems of poor specificity (there are multiple possible causal factors associated with each one) and incoherence (each DRF is effectively just a label for a wide variety of constructs including possible causes, contextual variables, and mental state factors). It is not clear whether they are best viewed as summary labels for predictor variables (i.e., they do not really exist as particular kinds of entities) or are latent constructs that causally generate their associated predictor indicators. Conceptual analysis is a core component of scientific inquiry at all its phases, but especially when deciding on the focus of inquiry problem and detecting phenomena. The use of ATOM will remind researchers to ask direct questions about the meaning of core concepts in a program of research. In addition, noting that DRF lack specificity and coherence is an excellent place to employ epistemic iteration to tease out the various conceptual strands embedded within the DRF categories and to formulate them with greater precision. For example, the general DRF of intimacy problems can be subdivided into emotional identification with children, lack of communication skills, lack of concern for others, social rejection, and so on. Emotional identification with children may be further divided into fear of adults, feeling like a child, viewing children as sources of emotional reward, seeing children as more accepting, being able to understand children, and so on (Thornton
2013; Ward and Fortune
2016). Refining DRF in this way may result in substantial theoretical progress and a richer description of the specific kinds of problems individuals who sexually offend against children exhibit in various domains. Approaching the analysis of DRF categories from a pluralist perspective means that there is a deliberate attempt to develop a range of contrasting perspectives and to explore to what extent they can interact in ways that improve understanding and practice.
Seeking for the “one true theory” and rejecting epistemic pluralism
In the correctional domain DRF are usually embedded within the RNR model of rehabilitation and have their theoretical grounding in Bonta and Andrews (
2017)
General Personality and Cognitive Social Learning perspective. There appears to be little willingness to acknowledge they can be reformulated in terms of agency constraints or protective factors (see Ward and Fortune
2016). Furthermore, the possibility of cultivating theory pluralism seems to be off the table, gauging from the lack of creative theoretical work on DRF (although see Thornton
2016). Approaching DRF with a comprehensive theory like ATOM, means that research will be broken down into distinct phases. First, there will be an attempt to detect the phenomena or “symptom” like features associated with DRF and their various components that constitute them will be modeled at different levels. For example, the DRF of self-regulation deficits has causal (e.g., lack of self-management skills), contextual (e.g., chaotic lifestyle), and mental state (e.g., feeling overwhelmed by emotions) features. Second, researchers will be encouraged to investigate the mechanisms that generate the symptomatic aspects of DRF and consider a range of possible explanations (e.g., differentiating between the cognitive and affective components of intimacy deficits). All of these distinct research problems and their respective cognitive tasks can reasonably be expected to result in a nuanced approach to theory building. In addition, the endorsement of theory pluralism should offset any tendency to prematurely reject alternative models in favor of a single unified explanation of DRF. From the viewpoint of theory pluralism there is no such thing as the “one true theory” of DRF. The use of epistemic iteration with its commitment to fallibilism should also help to combat a monistic orientation to explaining DRF.
Embracing impoverished theories of method
Some of the theoretical problems noted above appear to have their basis in the nesting of risk management research within inductivist or hypothetico-deductivist models of scientific method. Both of these theories of method privilege empirical research and place relatively little value on intensive theory construction and development (including conceptual, analysis—Ward
2014). However, because ATOM incorporates multiple types of inference (i.e., induction, deduction, abduction and causal reasoning) and makes room for related but distinct sets of scientific problems, both descriptive and explanatory work are considered legitimate scientific tasks. Thus, researchers will be expected to develop coherent conceptualizations of DRF and to situate these within detailed models of the mechanisms causing and constituting them. In addition, there will be a stress on careful analysis of DRF indicators in order to detect phenomena and test models of DRF. Theory pluralism will ensure researchers develop a range of models of DRF, while epistemic iteration coheres well with ATOM’s evolving multiphase model of inquiry.
Failure to distinguish distinct research tasks
The conflation of risk prediction and explanation evident in the literature on DRF is in part due to the adoption of theories of method that understate the importance of theoretical work. If researchers have a rich characterization of scientific problems and model of the inquiry process, then they will automatically explicate what specific cognitive task they are engaged in at any point in time. Without this guidance, it is easy to oversimplify matters and believe that attention to the factual basis of science will be sufficient to develop good scientific explanations and effective practice. According to ATOM, each of the phases of the inquiry process is characterized by a distinct suite of problems, which can be roughly divided into descriptive and explanatory phases. Addressing the problems at each of these general phases requires researcher to clearly state the question they are trying to answer. For example, preliminary questions could be: What do I mean by the concept of DRF? What are their key attributes? How should I break them down? Do they exist at different levels? And so on. In the explanatory phase, questions will revolve around the modeling process and their development and evaluation: What are the causes of DRF? What are they composed of? Would multilevel models provide a useful description of this DRF? Should I model them in functional (e.g., information processing stages) or mechanistic terms? Or both? What evaluation criteria should I prioritize: empirical adequacy, explanatory depth, coherence or heuristic value? And so on. Research problems phrased as questions will help researchers to be clear about exactly what tasks they are engaged in and to lessen the chances of them incorrectly shifting from the context of prediction to that of explanation. Epistemic iteration, with its stress on developing refined and theoretically rich categories from initially thin ones, also begins with some rather basic initial questions: What am I trying to do: explain, classify or predict? Where is my starting point? Finally, theory pluralism is much more relaxed about possible category mistakes with respect to models. It encourages the development of a variety of perspectives about a specific phenomenon or issue. This could mean that researchers intentionally attempt to create models that predict and explain, and also sharply separate them in another explanatory strategy. The aim is to build a wide range of explanatory models that lead to different empirical projects, and then to see which ones prove to be the most successful in solving the problems under consideration, in this case, crime.
Forensic classification
Scientific classification is usefully construed as the sorting of phenomena into non-arbitrary groups or sets of categories based on their core features or properties, for example, the types of symptoms associated with different kinds of diseases (Wilkins and Ebach
2014). This may occur prior to the development of explanatory theories although it is a conceptual task as it involves judgments about how certain phenomena should be grouped together. Wilkins and Ebach (
2014, p. 18) characterize classification in the following way: “To classify is to order the data, to find regularities even if you have no distinct idea why they cluster so. It is to find identity classes empirically, setting up some problem or thing in need of explanation.” The identification of scientific categories is important as they license researchers to make inferences about the entities in question (e.g., the course of a disease or behavior of a particular animal) and to predict outcomes based on an understanding of their central properties. Phenomena can be classified in different ways, depending on the goals and theoretical allegiances of researchers (Borsboom
2017; Wilkins and Ebach
2014). For example, in the area of psychopathology classification could be based on structural aspects of disorders (i.e., signs and symptoms) or in terms of their function (e.g., inability to meet specific types of need).
The classification of individuals on probation, in forensic settings or in prison is traditionally based on the types of harm they have caused to other people and the related actions that have led to these harms. Sorting individuals, the crimes they commit, and their crime-related problems into types is an important first step in the science and practice of forensic and correctional psychology. It helps researchers to concentrate on key problems and provides a common language for communicating their results to each other. Abstracting away irrelevant individual differences and concentrating on core properties shared by those convicted of different kinds of crimes makes it easier to describe their problems and ultimately to develop better explanatory theories. In the practice arena, treatment programs are typically developed according to types and subtypes of offences, for example, violent or sexual offending (or those who commit rape versus child molestation). In addition, clustering crime related problems into types based on factors such as kinds of crime, risk level, or dynamic risk factors enables practitioners to identify therapeutically useful treatment targets, and assign individuals to appropriate treatment programs and interventions (Bonta and Andrews
2017). Often the categories are used together, and different types of crime are subdivided according to risk level, with specific dynamic risk factors identifying the
types of changeable risk factors present. I will now briefly formulate criticisms of the practice of classifying individuals in terms of offense type, risk level, and DRF with respect to the six types of theoretical illiteracy outlined earlier. Alongside criticism, I will explore how utilizing the three strategies of pluralism, adopting a comprehensive theory of scientific method, and epistemic iteration can address them.
Uncritical and dogmatic acceptance of existing theories
Classification practices in forensic psychology have proceeded in the absence of explicit discussion of the role of classification in science (Ward and Carter
2019). In part this is due to the applied nature of forensic and correctional psychology and its stress on reducing reoffending rates. The theoretical legitimacy of basing the classification of offending related phenomena on properties such as type of crime, risk level or dynamic risk factors is simply assumed, not argued for (Ward and Carter
2019). To many it seems intuitively obvious that people who commit crimes such as fraud or violence do so because of their underlying antisocial dispositions; features that cluster together. It is assumed that classifying individuals according to the offences they commit, their level of risk, or number and type of dynamic risk factors captures all that is needed to effectively explain and reduce their chances of committing further crimes.
According a pluralist perspective, I think this is a mistake and that it is important to critically examine the use of classification systems based on crime in forensic and correctional practice and to consider alternative ways of classifying crime-related problems. For example, rather than focusing solely on the characteristics of individuals who harm others, concentrating on motivational systems, their associated goals, and the affordances (opportunities) provided by the environment provides greater understanding of the meaning of the individuals actions and how they are connected with non-offending aspects of their lives. That is, we ought to ask: What social and/or psychological tasks is the person engaged in? What outcome(s) are they implicitly or explicitly attempting to bring about (e.g., threat management, status enhancement, affiliation, care giving, mate seeking or resource acquisition—Ward and Carter
2019)? Adopting a comprehensive theory of method such as ATOM with its division of inquiry into descriptive and explanatory phases makes it easier to think about the different roles classification systems play in science, and also to explicitly consider a range of possible systems. Relatedly, in conjunction with pluralism, there is acceptance that different types of classification offer different things, such as those developed for prediction purposes (e.g., based on risk level) versus those intended to guided intervention (e.g., based on function or goals). Finally, the process of epistemic iteration can help researchers to grasp that the development of good classification systems can takes time and necessarily starts with fairly rudimentary categories. Both conceptual analyses and empirical research should be utilized to gradually develop a system, whether that ultimately entails its subdivision into different classification systems (e.g., offense types being replaced by risk bands) or the combination of different categories into an integrated classification system (e.g., risk levels supplemented by protective factors and DRF). For all of these tasks, it is necessary to engage in critical analyses.
View of science as strictly empirical in nature
Classifying people based on their kind of offence fails to subdivide them into meaningful psychological and behavioral categories. A significant issue—contrary to the assumptions of risk-oriented researchers and practitioners—is that what constitutes an offence (or type of offence) is normatively defined. The assumption made is that when criminal laws are broken, individuals and members of the community are significantly harmed, and that the types of crime (e.g., sexual offences and violent offences) specify the types of harm and their sources. The problem with incorporating normative concepts into correctional categories is that the categories themselves are derived with reference to social, political and legal institutions, meaning that treatment intervention based on these categories will not necessarily be targeting the causal features of the problematic behavior.
Implementing the strategy of model of explanatory pluralism when constructing different types of forensic classification systems makes it more likely that both the normative and factual aspects of classification will be taken into account. For example, classifying for treatment will be based on well-being related or prudential values as well as the level of psychological skill and knowledge possessed by individuals. Whereas, if the aim of a classification system is to predict the risk of reoffending then the guiding values will be ethical in nature (e.g., likelihood of harming others) and the factual basis related more to moral psychological features such as impulsivity or antisocial attitudes. With respect to the value of adopting a comprehensive theory of method, it is clear that problem formulation in science and theory evaluation rely on different types of values as well as knowledge-related considerations such as the depth of an explanation. A method such as ATOM has the theoretical resources to incorporate this diversity and to specify what kinds of values and factual issues are salient at different times of the classification construction process. Finally, epistemic iteration is based on normative and factual criteria as it is in part driven by researcher interests and in part by an awareness of the factual constraints in operation. For example, if the aim is to classify individuals for treatment, then an appreciation of the role of human needs in adaptive functioning is necessary and this concept is in part normative (i.e., what is essential for healthy functioning) and partly factual (i.e., determining what psychological and physical mechanism underlie need satisfaction).
Undervaluing the importance of conceptual analysis
Categories based on types of offenses are relatively
thin; the only property in common is the type of crime committed, for example, a sexual or violent offense. This problem arises from the heterogeneous nature of offenders, who often present with distinct and sometimes contrasting clinical profiles (Marshall et al.
2006). For example, one person may be socially unskilled, fearful of other people, and struggle to manage negative emotions while another could be socially high functioning and only possess a single problem such as deviant sexual interests (Mann et al.
2010). Assuming that persons who commit a particular
type of crime all manifest similar behaviors and difficulties reflects an overly simplistic view of the complex nature of human behavior, and fails to allocate offenders to intervention programs tailored to their individual needs (Taxman and Caudy
2015). Hence, classifying offenders based on their offence type does not provide meaningful categories as it does not highlight therapeutically useful treatment targets. We propose that offender classifications should focus on the cluster of problems exhibited by an offender, rather than on the offending behavior itself.
Furthermore, as stated earlier, the concept of DRF is a hybrid or composite one and thus theoretically incoherent. What we mean by theoretical incoherence is that DRFs are comprised of different kinds of conceptual categories and lack the integration required for use as explanatory constructs in research or practice contexts. However, somewhat ironically, the internal complexity of DRFs means that they are likely to be valuable predictors of re-offending (Ward
2016). Nevertheless, in their current conceptualization the role of DRFs in correctional classification is unwarranted and largely display poor practical utility.
Theoretical pluralism involves the construction of classification systems (a lower level theoretical task) of different types, at different levels of analysis. For example, self-regulation pathway models of reoffending (Ward and Hudson
1998) or typologies of violent offenders based on physiological indices such as heart rate variability (Raine et al.
2006). The development of each of these classification systems in part depends on the analyses of the key concepts involved, such as heart rate variability or offense pathways. Utilizing a comprehensive theory of scientific method such as ATOM is likely to remind researchers of the importance of addressing theoretical issues in the process of classification development. For example, identifying phenomena in the descriptive phase of investigation requires an analysis of data patterns and a rationale for deciding that a given pattern represents an aspect of the world such as a psychopathology symptom. The categorizing of symptoms into groups involves deeper theoretical analysis and complex judgements concerning their boundaries and relationships to underlying psychological mechanism and/or other symptoms. Finally, the strategy of epistemic iteration in the context of classification demands that researchers be clear about the meaning of initial categories and to justify any subsequent refinement and splitting or collapsing into higher level ones. Such clarity entails an awareness of the meaning (a conceptual task) of the relevant concepts in addition to the collection of statistical data.
Seeking the “one true theory” and rejecting epistemic pluralism
The allocation of offenders into
risk bands (low, moderate or high) fails to inform practitioners
which factors should be targeted in treatment, and
how those factors can be addressed. That is, although risk-prediction instruments can tell us who possesses the greatest risk of re-offending, they fail to provide any therapeutically useful information about those individuals. In addition, the subscales of risk assessment instruments such as the Level of Services Inventory‑R (e.g., attitudes/orientations, companions, accommodation—Bonta and Andrews
2017) measure state and surface features, which do not provide sufficient knowledge of the properties of problem behaviors. This leads to intervention plans that are inherently vague and nonspecific. In addition, because current classification systems grounded on offence type, risk level, and DRF were designed primarily for prediction purposes they do not offer much to policy makes and practitioners interested in resource allocation, treatment, and prison management. From this perspective the one “true theory” of classification is one oriented around risk prediction and management. It is simply assumed that a classification intended to ground risk prediction is suitable for treatment and other purposes. This is unlikely to be the case.
From what I said above it should be clear that given the competing interests of policy makers, police, researchers and practitioners, more than one classification system is likely to be required in forensic psychology. There is more than one way to “carve nature at its joints” in the criminal justice domain, depending on the questions you are asking: What is the likelihood someone will reoffend? What goals does an individuals’ offending serve? What tasks are they engaged in? How should scarce criminal justice resources be allocated? How can we best promote desistance from offending? Each of these questions demands its own classification system; a relevant way of clustering characteristics of individuals and their relationships to the environments that is capable of meeting its specific aims. In other words, it makes more sense to develop a plurality of classification systems rather than seek to find the one “true system” that will meet the goals of all criminal justice personal and stakeholders.
If you accept the above argument, it is obvious that pluralism is the best strategy for developing classification systems tailored to the interests of different criminal justice actors. A one size fits all approach will not work. A comprehensive theory of method is likely to assist in the process because of its appreciation that research problems drive scientific investigation, and reflect quite distinct questions. There is no such thing as the best research problem; they vary depending on the practical and theoretical problems confronting individuals and institutions. Finally, the strategy of epistemic iteration with its commitment to falliblism, category refinement, and innovation logically entails a commitment to the development of multiple classification systems if the interests and task requirements of an inquiry process require it.
Embracing impoverished theories of method and failure to distinguish distinct research tasks
I will only make a few brief comments on the final two examples of theoretical illiteracy as what I have argued earlier is applicable in both instances. In short, reliance on restricted theories of method is likely to result in poor classification systems, primarily because of their neglect of conceptual and theoretical work. Given that I have argued that the construction of classification schemes is in part a theoretical task, it follows that any theory of method that ignores or minimizes the role of theory will lead to the other kind of illiteracy problems documented above. Furthermore, the endorsement of an impoverished theory of method means that the full range of distinct cognitive tasks associated with classification (and science) will be overlooked and there will be a significant risk of mistakenly thinking that one classification system can be equally valuable in the contexts of prediction, explanation, treatment planning and resource allocation. Adopting the three strategies of model pluralism, epistemic iteration, and following a comprehensive approach to inquiry, should minimizes the chances of this occurring. For example, understanding that description precedes explanation in science, and that there is more than one reasonable way to describe or group phenomena together, should encourage researchers to be explicit about what task they are engaged in and what problems they want a particular classification system to address (e.g., treatment planning versus risk prediction). Thus it is less likely that cognitive tasks of prediction, treatment planning, and explanation will be conflated. In turn, this should create a more dynamic and pluralistic scientific culture where competing ideas and different solutions to problems are actively encouraged.