Socioeconomic marginality in sentencing: The built-in bias in risk assessment tools and the reproduction of social inequality

Introduction

The use of actuarial risk assessment tools in sentencing decisions has become common practice in various jurisdictions such as several states in the US, Canada, the Netherlands and the UK (e.g. Fitzgibbon, 2008; Hannah-Moffat, 2013; Harcourt, 2007; Monahan and Skeem, 2015; Starr, 2015; van Wingerden et al., 2016). Criminologists and legal scholars have criticized risk assessment tools for their potential bias against racial/ethnic minorities and women, which could result in sentencing disparities (e.g. Chenane et al., 2015; Hannah-Moffat, 2009, 2016; Hannah-Moffat and Maurutto, 2010; Harcourt, 2007; Holtfreter and Cupp, 2007; Monahan, 2006; Petersilia and Turner, 1987; Skeem and Lowenkamp, 2016; Skeem et al., 2016; Smykla, 1986; Tonry, 1987, 2014; Whiteacre, 2006). While many scholars have pointed to the role of assessing socioeconomic factors in producing ethnic/racial and gender bias, relatively little attention has been given to the problem of socioeconomic bias in itself (but see Goddard and Myers, 2016; Starr, 2014, 2015; Tonry, 2014; more below). Widely used risk assessment tools such as the Level of Service Inventory-Revised (LSI-R) (used in Canada and the US), the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) (US), the Offender Assessment System (OASys) (UK) and the Recidive InschattingsSchalen (RISc) (the Netherlands), which are applied to general offender populations, include socioeconomic factors such as education, employment, financial situation and accommodation. This is problematic because, put simply, socioeconomic marginality contributes to a higher risk score, which in turn could translate into more severe sentences for these individuals (e.g. incarceration instead of a community sanction, or a longer sentence) compared to their more privileged counterparts. Early critics of risk assessment have pointed out that socioeconomic factors such as employment and education are ‘class-based variables that, in effect, discriminate against the poor’ (Smykla, 1986: 130–131) and that such factors ‘should be forbidden’ in sentencing criteria (Tonry, 1987: 408). More recently, Starr (2014, 2015) has offered an elaborate critique of ‘punishing based on poverty’ and the ‘scientific rationalization of discrimination’, while Goddard and Myers (2016) have described risk-based assessment as ‘evidence-based oppression’ of marginalized youth.

This largely conceptual article builds and elaborates on these critiques in order to develop a sociological analysis that focuses more explicitly on how including socioeconomic marginality as a risk factor in sentencing decisions reproduces – and possible exacerbates – disparities in sentencing as well as social inequality more generally. Using the sociological concept of ‘cultural process’ (Lamont et al., 2014), I conceptualize sentencing based on risk assessment as a process through which resources and recognition are distributed, based on meaning-making through identification and rationalization. In the next section I briefly discuss the role of socioeconomic factors in risk assessment tools. I then introduce the concept of ‘cultural process’ and examine how risk-based sentencing is likely to reinforce disparities and inequalities as it works to distribute opportunities and resources differently among convicted individuals based on their socioeconomic status. The article ends with outlining several directions for further research and debate.

The practice and problem of including socioeconomic marginality in risk assessment tools

Including socioeconomic factors in assessment tools is by no means a recent development. For example, the first American risk assessment tool developed by Burgess in the 1920s included ‘work record’ (for an overview of early risk assessment tools, see Oleson, 2011). So-called second-generation tools such as the Violence Risk Appraisal Guide (VRAG) and Static-99, typically only include static factors – criminal history and certain demographic characteristics such as age – that may change over the life course but cannot be targeted through intervention (for an overview, see Hamilton, 2015). It is with the development of third-generation ‘risk/needs’ assessment tools that aim to address ‘dynamic risk factors’ or ‘criminogenic needs’ that socioeconomic factors were reintroduced (Andrews and Bonta, 2010). The most well-known example probably is the LSI-R, developed in the 1980s in Canada (Andrews and Bonta, 2010; for a detailed history, see Maurutto and Hannah-Moffat, 2006). The LSI-R is also the most commonly used tool in American jurisdictions (Hamilton, 2015; Harcourt, 2007). The tool is based on Andrews and Bonta’s (2010) ‘psychology of criminal conduct’ theory and assesses the ‘Central Eight’ domains that are related to reoffending, among which is ‘social achievement’ (initially based on Gendreau et al., 1996). The tool consists of 54 items of which 10 items that measure education and employment (e.g. ‘currently employed’, ‘never employed for a full year’, ‘ever fired’, ‘less than grade 10/12’), two items that measure financial situation (‘financial problems’, ‘reliance on social assistance’) and three items that measure accommodation (e.g. ‘unsatisfactory housing’, ‘residential stability’) (Caudy et al., 2013; Hamilton, 2015).

Comparable risk/needs assessment tools are the OASys which is used in the UK (Robinson, 2003), the RISc (translates as Recidivism Assessment Scales) in the Netherlands (van Wingerden et al., 2016) and the COMPAS in several American states (Hamilton, 2015; New York State (NYS), 2015). All three tools inform treatment as well as sentencing and parole decisions and include items that measure education/employment, financial situation and accommodation (Fitzgibbon, 2008; NYS, 2015; van Wingerden et al., 2016). For example, the RISc consists of 62 items in total, divided among 12 scales (3–8 items per scale) such as ‘accommodation’ (4 items), ‘education, work and training’ (7 items) and ‘financial management and income’ (4 items); specific items include ‘problematic employment history’, ‘unemployed/unable to work’, ‘work experience’, ‘lack of work skills’, ‘no education/diploma’, ‘depends on others for income’, ‘financial problems’, ‘debts’, ‘history of homelessness’, and ‘no suitable/permanent accommodation’ (van Wingerden et al., 2016).

Socioeconomic factors can make up 10 to 25 per cent of the total number of items assessed. For example, socioeconomic factors are assessed in 15 of 62 items in total in the RISc, 15 of 54 items in the LSI-R; and add up to 10 of 100 points in the OASys Violence Predictor (OVP), a variation of OASys for violent offenders (Howard and Dixon, 2013). Their exact contribution to the overall risk score may differ when items are weighted, which is not always transparent as sometimes algorithms are proprietary (e.g. COMPAS). Furthermore, some of the other items, such as those assessing leisure activities, seem correlated with socioeconomic status (see Goddard and Myers, 2016). But even if socioeconomic factors would contribute little to the overall risk score, they could and do make the difference between assessing an individual as ‘low’ or ‘moderate’ risk, or ‘moderate’ or ‘high’ risk (Starr, 2014, 2015). Thus, that socioeconomic factors are among many factors assessed cannot be an argument to disregard potential social consequences. Regardless of how other items are scored, underprivileged individuals are confronted with a higher risk score and, consequently, are more likely to face a custodial sentence or to face a longer custodial sentence, compared to their more privileged counterparts.

A critical evaluation of socioeconomic factors in risk assessment tools is essential given concerns among scholars as well as criminal justice actors about socioeconomic disparities in sentencing (Holder, 2014; Reiman and Leighton, 2016; van Wingerden et al., 2016; Western, 2006). As Starr notes, basing risk assessment on socioeconomic factors means using ‘dry, technical language to obscure discrimination that we would otherwise never accept’ (2015: 229). Furthermore, ‘if judges or policymakers would be embarrassed to embrace ideas like “we should increase people’s sentences for being poor” openly, then they should not do so covertly by relying on a risk score that is substantially driven by such factors’ (Starr, 2015). ‘Covert’ here should not be read as meaning ‘biased interpretation’ of items that turn out to disadvantage certain social categories. To the contrary, and unlike racial/ethnic and gender biases, socioeconomic bias results from direct discrimination on the basis of socioeconomic marginality: it is a built-in bias. ‘Covert’ in Starr’s remark rather refers to the fact that many of the people involved (defendants, lawyers), let alone the general public, are not aware that risk assessment instruments inform sentencing decisions, or they do not know which and how individual characteristics are assessed (Starr, 2015). Nonetheless, it may well be, given the widespread support for current risk assessment tools, also among judges (e.g. Kopf, 2015; Wolff, 2008), that judges and policymakers would not be embarrassed at all to defend sentencing based on socioeconomic status, as long as it can be maintained that doing so contributes to public safety.

It could be argued that risk assessment intends to discriminate (in the sense of ‘differentiate’) based on socioeconomic status because socioeconomic factors have predictive validity: research shows that problems related to employment, educational, finances and housing are criminogenic factors and, more specifically, that they predict reoffending (a widely cited source is Gendreau et al., 1996; specific evaluations of the predictive validity of socioeconomic factors include Andrews and Bonta, 2010; Howard and Dixon, 2013; Johnson et al., 2011; Raynor et al., 2000). If socioeconomic factors improve predictions of reoffending, it could be argued that it is irresponsible, in terms of public safety, to not include such factors in sentencing decisions. There is an argument to be made that the results of validity studies are inconclusive (e.g. Austin et al., 2003; Brennan et al., 2009; Caudy et al., 2013; Petersilia and Turner, 1987). But rather than debating the predictive validity, I wish to draw attention to the consequences of assessing socioeconomic marginality as risk factor. Notwithstanding the fact that the evidence indeed points towards predictive validity, an evaluation of the outcomes requires that we take into account both intended (public safety, efficiency) and unintended outcomes (disparities, social inequality). In the 1980s, the use of factors such as education, vocational skills, employment record, income and residential stability was generally seen as ‘inappropriate’ for including in sentencing decisions by American courts and parole boards (Tonry, 1987). Both the Minnesota sentencing guidelines (1980) and the Sentencing Reform Act (1984) stated that such factors could not be used in sentencing decisions due to their ‘socially skewed impact’ (Tonry, 1987: 398). Similarly, for race and race-related items, it has been decided that discrimination and racial disparities are not acceptable, and thus race and race-related items have been removed from predictive instruments (Harcourt, 2007; Tonry, 2014). Evaluating the legitimacy of practices based on criteria such as validity and evidence base thus is too limited, as it pushes aside moral questions and brushes over social-political, economic and cultural conditions that shape socioeconomic marginality (cf. Goddard and Myers, 2016; Hannah-Moffat, 2016; Silver, 2000). Rather, we need to evaluate risk assessment in its moral and social context.

Furthermore, analysing the potential discriminatory effects of risk assessment has broader relevance. First, research in various countries has pointed to socioeconomic and racial disparities in sentencing (e.g. Kutateladze et al., 2014; Millie et al., 2007; Reiman and Leighton, 2016; van Wingerden et al., 2016) and risk assessment based on socioeconomic factors could play a role in how such disparities emerge. Second, there is increasing academic interest, particularly related to mass incarceration in the US, in understanding how punishment, especially imprisonment, might reinforce socioeconomic inequality more generally, considering the adverse effects of prison sentences on education, work and income (Pager, 2008; Wakefield and Uggen, 2010; Western, 2006; Western and Pettit, 2010). Such adverse effects may also exist in less unequal and less punitive societies, such as the Netherlands (Ramakers et al., 2014). It is difficult to determine whether certain sentencing practices merely reflect existing inequalities, or whether such practices in themselves create or reinforce inequalities, because offender populations are disproportionally often socioeconomically marginalized and disadvantaged (Loeffler, 2013; Ramakers et al., 2015; Wakefield and Uggen, 2010). One way to gain more insight into the ways in which sentencing disparities emerge and in how sentencing might in itself produce inequality, is to trace in more detail the processes through which sentences are imposed and how this impacts opportunities and resources. This article aims to do so by zooming in on the role of risk assessment.

It should be noted that risk/needs assessment tools may be used for multiple purposes, ranging from informing sentencing and parole decisions to guiding decisions on treatment and supervision (Hannah-Moffat, 2005; Monahan and Skeem, 2015). This article focuses on the role of risk assessment in sentencing decisions. It is particularly in sentencing decisions that including socioeconomic factors is problematic – more so than in decisions about supervision and treatment (cf. Starr, 2014). In reality, however, the goals of sentencing and treatment cannot always be separated, as judges may impose treatment conditions as part of a sentence. I return to the intertwining of risk and needs assessment, and sentencing and treatment later in this article. For analytical purposes I simplify the decision-making process and focus on the potentially differentiating effect of a convicted individual’s risk score in two types of decisions: (1) whether or not to impose a custodial sentence (jail or prison), as opposed to diversion (a community sentence, treatment or fine) – as is for example explicitly the goal in the state Virginia (Kleiman et al., 2007) – and (2) determining the duration of a custodial sentence – including eligibility for parole, which in effect is also a decision about the duration of imprisonment (Harcourt, 2007). In the following section I examine in more detail how risk assessment could reinforce disparities and inequality.

Tracing the cultural processes that produce social inequality

Statistics on the incarcerated population in the US show that socioeconomic disparities have increased significantly between 1980 and 2010: ‘the spectacular growth in the American penal system over the last three decades was concentrated in a small segment of the population, among young minority men with very low levels of education’ (Western and Pettit, 2010: 18). But also in countries with lower levels of inequality and lower incarceration rates, the population that is under control of the criminal justice system tends to be poorer, lower skilled and more often homeless, compared to the general population. In the Netherlands, for instance, the labour market position of convicted individuals prior to their imprisonment is already weak: ‘Starting with a low educational attainment, their subsequent employment career is characterized by long periods of unemployment, “off-the-books” employment, dismissals and job shifts’ (Ramakers et al., 2014: 65). Starr (2014, 2015) has – in her elaborate critique on ‘evidence-based sentencing’ in the US – made the point that including socioeconomic factors in risk assessment will increase sentencing disparities rather than reduce them. Risk assessment classifies as ‘higher risk’ exactly those social categories that are already overrepresented in the criminal justice system: the un- or low-skilled, the unemployed and the unhoused. When sentencing decisions, especially decisions about whether or not to incarcerate someone, are partly based on evaluations of socioeconomic status, as is done through risk assessment, the link between sentencing and socioeconomic status is reinforced. Harcourt (2007) has called this the ‘ratchet effect’: prediction instruments and profiling practices increase the disproportionality between the composition of the actual offender population and the population that is in some way under control of the criminal justice system. Harcourt stresses that it is important to recognize that this effect is produced even if the relationship between socioeconomic marginality and reoffending is real (i.e. not an artefact of criminal justice policies and practices). Whether ‘evidence based’ or not, a greater focus on particular groups – more policing, more prosecution, more imprisonment, longer sentences – means that their share in the total population that is under control of the criminal justice system will grow, and continue to grow, and that disparities will increase (Harcourt, 2007).

Moreover, when sentencing affects an individual’s opportunities in life, it also shapes social inequality in societies more generally. Following Fraser (1995), we can broadly define social inequality as the unequal distribution of (access to) resources that matter for people’s quality of life and wellbeing, including material and immaterial resources as well as recognition (cf. Lamont et al., 2014). Various studies have demonstrated the adverse effects of (length of) imprisonment on educational achievement, employment opportunities, income and housing (e.g. Apel and Sweeten, 2010; Pager, 2008; Ramakers et al., 2014; Western, 2006). In this way, we can understand sentencing – and criminal-justice decisions more generally – as a social process through which resources are distributed and, consequently, social inequality is either reduced or reinforced. Even if adverse effects are largely selection effects (because the offender population is already underprivileged, see Loeffler, 2013), it seems safe to say, based on research, that imprisonment makes it difficult for individuals to maintain, let alone improve, the quality of their life and wellbeing (e.g. Dirkzwager et al., 2014). The ‘best case’ scenario is thus that sentencing based on biased risk assessment merely reproduces inequality; in the worst case it exacerbates inequality.

To better understand how this would work, I draw on the work of Lamont et al. (2014) to conceptualize sentencing based on risk assessment as a decision process that is grounded in two ‘cultural processes’: identification and rationalization. Cultural processes link cognitive categorizations in people’s minds, on the one hand, and the distribution of important resources and recognition, on the other. Put differently, through ‘inter-subjective meaning-making’ people create shared classification systems that ‘sort out’ people and actions (Lamont et al., 2014: 582). When classification systems inform actual decisions that impact people’s life, the sorting out of people is translated into the (unequal) distribution of resources among categories of people, thus shaping access to resources and hierarchies of status and worth. It is in particular the state, Lamont et al. (2014: 585) argue, that ‘wields immense power in shaping and legitimizing systems of categorization’ through designing and executing laws and social programs. They further stress that the production of inequality does not have to be intentional; it could be an unintended consequence of other, intentional actions. Without making any assumptions about punitive sentiments – either of governments or of the general public, or both – towards socially marginalized groups, the analytical approach proposed by Lamont and colleagues is to trace the processes through which resources and recognition are distributed. In this way, we can shed new light on the relation between sentencing, particularly imprisonment and social inequality.

Lamont et al. (2014) have theorized that ‘identification’ and ‘rationalization’ are two broad types of cultural processes that are central in producing inequality. Identification is ‘the process through which individuals and groups identify themselves, and are identified by others, as members of a larger collective’; stigmatization, or the negative qualification of identities, is an example (Lamont et al., 2014: 587). Rationalization, famously described by Max Weber, broadly refers to ‘the displacement of tradition and values as motivation for action by a means-end orientation’, for example through standardization and evaluation (Lamont et al., 2014: 591). Both processes are central in risk assessment based on socioeconomic marginality (cf. Silver, 2000). Risk assessment evaluates socioeconomic marginality negatively, first by the mere fact that it is included as a risk factor, second by prescribing that a higher overall risk factor is interpreted to mean a greater danger to the public and thus should translate into a higher sentence. Both Harcourt (2007) and Starr (2014, 2015) have pointed to the symbolic message that is associated with disproportionate sentencing based on certain offender characteristics. Starr is concerned with treating poverty as a risk factor, which ‘is also endorsing a message: that it considers certain groups of people dangerous because of who they are, not what they have done’ (2015: 230). Profiling based on prediction about who will (re)offend comes with a social cost, Harcourt (2007) argues, because it tends to create unintended stigma that attaches to, in this case, the unemployed, unskilled and unhoused. Incarceration or longer sentences for offender categories who are assessed as higher risk because of their socioeconomic status reinforces the negative identification of socioeconomic marginality. Through defining which social categories are worthy of beneficial policies and which categories deserve punitive policies, policymakers allocate resources (Gans, 1995). Public policies in turn send messages about which categories are deserving and what are appropriate kinds of attitudes towards them (Schneider and Ingram, 1993). Deserving citizens are supported, while undeserving citizens tend to be punished. In sentencing decisions this means: socioeconomically ‘integrated’ or ‘productive’ offender categories – those who are skilled, employed or employable, and with a stable residence – are more deserving of diversion away from imprisonment, so that they can maintain their socioeconomic status. Socioeconomically marginal offender categories, on the other hand, are undeserving of diversion and, consequentially, of the opportunity to improve their socioeconomic status, as their options to develop a mainstream way of life are hindered during and after imprisonment. In addition, when risk assessment is legitimized based on the supposed benefits of diversion for low-risk individuals, while the costs of non-diversion for high-risk individuals (i.e. incarceration and all its consequences) are ignored, the message is reinforced that convicted individuals who are also underprivileged are less worthy of opportunities to improve their life.

What is more, risk assessment reinforces the notion that socioeconomic marginality is the sole responsibility of the individual (Goddard and Myers, 2016; Hannah-Moffat, 2016). This is not surprising, given that current risk assessment tools are grounded in a social-psychological theory of offending that looks for explanations related to the individual rather than to society (Andrews and Bonta, 2010). But it does more. Andrew and Bonta’s argument for including socioeconomic achievement in the LSI-R tool on the ground that it is an ‘achieved’ status, as opposed to ‘social class’ (family background) which would be an ‘ascribed status’, is consistent with the dominant cultural narrative in contemporary Western societies that we live in ‘classless’ societies in which people shape their own life course (e.g. Beck, 1992; Lamont, 2000; Savage, 2000; van Eijk, 2013). However, research in Europe and the US shows that opportunity structures are less open than we would expect in classless societies: people’s educational and income level still depend on the socioeconomic status of their parents (Causa and Johansson, 2009; Putnam, 2015). In addition, structural factors such as rising inequality hamper intergenerational social mobility (Corak, 2013). Socioeconomic opportunities more generally are also affected by welfare reforms, economic crises and stagnating wages for low- and mid-level jobs, which have led to a growing number of people, including those who are working, living in poverty. By disregarding the social and historical context of the marginal positions of social groups, risk assessment individualizes socioeconomic marginality and thus shifts all responsibilities to individuals not only for the crime they have committed but also for their socioeconomic status (Goddard and Myers, 2016; Hannah-Moffat, 2016; see also Hannah-Moffat and Maurutto, 2010, on the individualization of gendered and racial inequality). The individualization of class is in a specific way problematic for individuals who come into contact with the criminal justice system repeatedly, as social structures can enable as well as constrain possibilities for agency in desistance processes (Farrall et al., 2010; Hannah-Moffat, 2016). For example, the transition to knowledge economies has meant fewer low-skilled jobs and a reconfiguration of ‘employability’, while the recent financial crisis has impacted overall unemployment rates; housing and urban policies have resulted in a smaller stock of affordable housing; and increasingly punitive criminal policies have meant fewer resources to support convicted and formerly incarcerated individuals with ‘reintegrating into the mainstream society’ in favour of managing risky populations (Farrall et al., 2010). A final structural barrier is the way in which social institutions ranging from housing associations and private landlords to educational institutions and employers manage risks by excluding people who have a criminal record (e.g. Carey, 2004; Pager, 2008). For individuals who cycle in and out of the criminal justice system, assessing socioeconomic factors could mean that they get caught up in a vicious cycle of being punished more severely due to a higher risk score based on socioeconomic marginality, which in turn results in limited opportunities to get back on track, which in case of a repeated conviction and assessment results in a higher risk score, and so on and so forth. In this way, risk policies combined with risk assessment create double punishments for repeat offenders, and in this way makes them even more marginalized (Goddard and Myers, 2016).

From a legal-philosophical standpoint, Tonry (1987, 2014) has argued that including socioeconomic factors violates the ethical proposition that ascribed characteristics for which individuals bear no responsibility, such as race, ethnicity, gender and age, should not play a role in parole and sentencing decisions. Similarly, the former Attorney General of the US Justice Department has argued that sentencing ‘should not be based on unchangeable factors that a person cannot control’, e.g. the defendant’s education level, socioeconomic background and neighbourhood (Holder, 2014, italics in original). Moreover, Tonry (2014) has argued that it is unethical and unjust to punish people for making personal choices about education, work and living arrangements, especially in societies that highly values freedom and personal autonomy. In other words, these choices are not blameworthy in the context of a criminal act (cf. Monahan, 2006). Arguing against including socioeconomic factors by saying it is a legitimate choice for people to make seems contrary to arguing that socioeconomic status is outside one’s control. However, it fits perfectly within a perspective that ignores the role of structural factors, which logically leads us to holding individuals responsible for making the ‘wrong choices’ and thus for their marginality. The fact that policymakers and criminal justice actors do not object to punishing individuals either for lawful personal choices or for factors that are out of their control, expresses dominant cultural narrative that socioeconomic factors are thought to be ‘under the control of the offenders and therefore expressions of their inclinations and values’ and therefore are of ‘moral significance’ (Moore, cited in Monahan, 2006: 398). Individuals are made responsible for their own socioeconomic status, and thus seen as blameworthy for not conforming to a mainstream way of life.

Proponents might argue that risk assessment will not stigmatize socioeconomic marginality because risk assessment tools cannot specify the causes of criminal behaviour. In fact, Monahan and Skeem (2015) argue that conflating risk and blame in risk assessment is a ‘category error’: blame and risk play a role in different types of decisions – that is, first conviction and then sentencing. However, in practice it would be difficult to keep separate judgements about blame and risk. Monahan and Skeem have further argued that technically it does not matter whether risk factors are causal or not, if the goal is not to understand but solely to predict behaviour. Hence, most factors in risk assessment, especially socioeconomic factors, are rather ‘variable risk factors’: studies have demonstrated their statistical correlation with recidivism, and they precede recidivism in time, and while they theoretically can be changed to reduce risk it has not been empirically determined that they do (Monahan and Skeem, 2015). More precisely, due to a lack of rigorous testing in randomized control trials it cannot be determined that they are causal risk factors (Monahan and Skeem, 2015). It is possible, and not unlikely, that the relation between socioeconomic factors and reoffending is causal for some individuals, but spurious or indirect for others. That is, third factors could cause both re-offending and problems related to education, employment, income and housing. For example, criminal history is not only predictive of reoffending, but it also makes it more difficult to get a job, degree or housing, as many employers, colleges and housing providers exclude people who have a criminal record. Offending might follow from substance abuse, criminal thinking patterns or anti-social personality – all factors that according to Monahan and Skeem (2015) have been identified as causal factors – but each of these factors may also cause socioeconomic problems. Even having ‘antisocial associates’, and thus limited sources of social capital and encouragement, could affect employment and education in addition to enabling criminal behaviour.

However, denying that risk assessment makes any claims to causality between socioeconomic marginality and offending conflicts with the claim that advanced risk/needs assessment tools would be superior exactly because they are theory-driven (Andrews and Bonta, 2010). After all, Andrews and Bonta, as did the developers of the COMPAS, set out to address criminogenic factors. But even if the instruments would be purely data-driven, it is a rather technical argument that draws attention away from the social context in which risk assessment operates and the way in which people attach meaning to it. The scientific language and confusing terminology, especially around ‘dynamic risk factors’ which are also often called ‘criminogenic needs’, make it difficult for people, including legal professionals, to understand that very few items, if any, included in risk assessment tools have been proved to cause criminal behaviour or reoffending. Hence, the danger of ‘slippage between risk prediction and individual causation’ is real (Hannah-Moffat, 2013: 273). Even among practitioners, who are generally not trained in statistics but are trained in interpreting these tools, correlation tends to become causation (Hannah-Moffat, 2013). Slippages between correlation and causation are frequently made in professional descriptions of risk assessment. For example, the OASys manual states that it is used to ‘help assessors in understanding the “why” of offending’ (National Offender Management Service (NOMS), 2009: 8) and the COMPAS manual details the criminological theories that ‘help us understand more about why people make their behavioural choices’ (Northpointe, 2012: 6). An additional complication follows from the merging of risk and needs in current ‘risk/need assessment tools’. Third- and fourth-generation tools are seen as superior to second-generation tools, because they include not only static factors (e.g. criminal history) but also ‘dynamic risk factors’ (e.g. socioeconomic factors), which are also often called ‘criminogenic needs’ (Baird, 2009; Hannah-Moffat, 2005). However, as Baird points out, while ‘the mere existence of a need does not always mean it is “criminogenic”’, labelling needs as criminogenic ‘[implies] a claim about causality that generally far exceeds what can legitimately be concluded from the assessment data’ (2009: 9). If anything, including socioeconomic factors in risk assessment tools thus is likely to convey the message that socioeconomic marginality causes reoffending, thereby attaching a stigma not only to marginalized offenders but also to socioeconomic marginality in itself. Risk assessment involves a process of negative identification of underprivileged offender categories by classifying them as higher risk, thus justifying higher sentences, which consequently impacts convicted individuals’ opportunities to maintain or improve their quality of life and wellbeing.

The cultural process of negative identification, or stigmatization, as described in detail in the above, is accompanied by a process of rationalization (cf. Silver, 2000). While Lamont and colleagues seem to theorize that rationalization could in itself produce inequality, it might be more plausible to follow Silver’s (2000) analysis that rationalization works to conceal the ways in which criminal justice policies and practices engage in social classification. Of particular importance is that standardized decision-making is seen to make decisions ‘neutral’ and ‘fair’ (Lamont et al., 2014: 591). More concretely, the rational, scientific and ‘value-neutral’ approach to sentencing works to obscure the ways in which risk assessment attaches a stigma to certain individual characteristics (Silver, 2000). Put differently, the stigmatization of socioeconomic marginality is accomplished ‘under the ideological banners of reason and rationality’ (Silver, 2000: 125). The social sciences, Silver argues, contribute to the rational sorting out of deserving and undeserving offender categories in at least two ways. First, by advancing technologies for managing populations – in this case by developing, testing and advancing risk assessment tools. Second, by providing an interpretive framework which justifies population management – here we can think in particular of the frameworks of ‘evidence-based practices’ and ‘what works’, the criminological theories that inform risk assessment (notably Andrews and Bonta’s (2010) social-psychological theory), and the focus on predictive validity and general applicability in evaluating risk assessment tools. Fundamental ethical questions (e.g. regarding bias, individual responsibility) are neglected as scientific rigor becomes the ethical criterion for judging risk assessment tools (Silver, 2000: 139). The emphasis on validity in scholarly evaluation of risk assessment, rather than on ethical and social criteria for including or excluding certain items, proves the dominance of the ‘evidence based’ and ‘what works’ frames.

The use of technical language not merely obscures discrimination, as Starr (2015) argues, but neutralizes or ‘launders’ it: outcomes that are related to bias are later taken as neutral measures of offending (Ward, 2015). Another ‘bias-laundering routine’ (Ward, 2015) is the depiction of risk assessment as ‘objective’. The benefits of actuarial assessment are usually pitted against unstructured clinical assessment which would have not only poor predictive accuracy but high discriminatory potential as well (Andrews and Bonta, 2010; Hannah-Moffat et al., 2009). Following this argument, it is discretion that gives room to prejudice and biased decision-making, while actuarial risk assessment structures decision-making and thus limits discretion and therefore is ‘objective’ and ‘truthful’ (Hannah-Moffat et al., 2009: 399). Some proponents thus argue that risk assessment tools are a solution to subjective and biased decision-making, such as decisions about sentencing. For example, a manual for prison and probation officers in the UK states that risk management is ‘fair and just – justified, non-discriminatory, does not overintrude on particular groups and that risk management should be ‘SMARTA’ – SMART with an additional A for ‘anti-discriminatory’ (NOMS, 2009: 9). I set aside here the observation of some scholars that in reality structured risk assessment may rather be an ‘actuarial illusion’ (Hannah-Moffat et al., 2009), as risk scores can and are overruled by practitioners when their professional judgement deviates from the risk assessment outcome, for example when they think that marginality should not result in a higher score (see also Ansbro, 2010). The point is that objectivity is similarly an illusion, as biased or prejudiced decision-making is merely replaced with a standardized procedure that has a built-in bias against socioeconomic marginality. In summary, rationalization works to legitimize stigmatization and thus the reproduction of marginality and inequality through risk-based sentencing procedures.

Towards critical assessment and debate

Thirty years ago, Tonry observed that ‘almost everyone who writes about predictive sentencing seems uncomfortable with it’ (1987: 400). That certainly has changed. In general, risk assessment is now seen as a solution to many problems, even mass incarceration. Many scholars, also those who raise questions about potential biases, engage in improving risk assessment tools rather than evaluating them within a moral and social context (cf. Silver, 2000). Furthermore, while the practice of including socioeconomic factors in risk assessment tools has been scrutinized in relation to racial/ethnic and gender bias, much less has been said about socioeconomic bias as a problem in itself. Harcourt (2007) suggests that as individual traits are deemed to be more closely related to criminal behaviour, the less problematic they are deemed in terms of discrimination. This might be why racial/ethnic bias is viewed as more troublesome than socioeconomic bias: there seems to be consensus among criminal justice actors, as well as among many criminologists who develop, test and advocate for risk assessment tools, that socioeconomic factors are related to reoffending – and offending more generally – and that this justifies including them in risk assessment and sentencing decisions, so as to contribute to public safety. However, this should in my view not be a reason to forego a critical assessment of whether or not to include socioeconomic factors, especially given that individuals who come before a judge often are underprivileged already. A built-in socioeconomic bias in risk assessment tools will contribute to and produce disparities in sentencing. It is therefore remarkable, as Starr (2014) has noted, that in the US risk assessment has widespread support as a solution to mass incarceration: if risk assessment is to reduce mass incarceration, Starr argues, it should single out marginalized groups for diversion, instead of punishing them more severely. In the US, where calls for reforming the criminal justice system, spurred by concerns about the costs (financial and social) of mass incarceration, are growing louder, there is momentum to introduce policies and practices that avoid the devastating impact on socioeconomically marginalized communities (as well as communities of colour). But the problem is not limited to the US. Also in other countries, risk assessment is promoted because of its promise of fair and equal decision-making. Although in other countries the call to reform sentencing policies and practices may be less urgent, risk assessment nonetheless is welcomed as a way to cope with diminished budgets and resources as well as ongoing quests to protect the public from offenders. With this article, I hope to have made clear that, despite claims of objectivity and fairness, risk assessment tools in their current form are not unbiased. In addition to potential racial/ethnic and gender bias, the built-in socioeconomic bias is problematic in itself, for it produces disparities in sentencing. Moreover, sentencing that is based on the socioeconomic status of individuals might work as a process that reproduces and perhaps exacerbates social inequalities in quality of life, opportunities and wellbeing more generally. Particularly given longstanding concerns about socioeconomic and racial disparities in criminal justice decision-making, an evaluation of risk assessment should take into account not only public safety but also unintended societal consequences.

An analysis of the role of risk assessment in sentencing is not only relevant for those studying, evaluating and advancing risk assessment tools, but also for scholars who have argued that imprisonment produces social inequalities. They have generally looked at the consequences of prison sentences for individuals’ life course, particularly opportunities for educational achievement, employment and income. Risk assessment in its current form may have a contributing role in producing such adverse effects of imprisonment. For research on the relation between imprisonment and social inequality, it may thus be valuable to zoom in on specific components of decision-making processes. The problem of selection effects makes it difficult to determine whether unequal opportunities after sentencing are related to sentencing itself. A sociological analytical focus on social or cultural processes could be of additional help in unravelling the relationship between sentencing practices on the one hand, and social inequality on the other.

To more fully understand the role of risk assessment in the (unequal) distribution of resources, opportunities and recognition, we should consider the practice of risk/needs assessment in its entirety. Needs assessment aims to do exactly the opposite from what risk assessment does: allocating services and thus (access to) resources to high-risk offender categories. The identification of social categories through assessment could benefit underprivileged groups, when identification and rationalization result in distributing resources towards socioeconomically marginalized groups, as opposed to hindering access to resources. In this way, needs assessment might be a cultural process that works to decrease social inequality (provided that budgets are plentiful and that convicted individuals get the services they need). However, the outcomes of both risk assessment (produces inequality) and needs assessment (decreases inequality) may conflict with each other to such extent that the rehabilitative efforts are undermined when they are accompanied, or preceded, by sentencing based on risk assessment. If risk assessment in sentencing produces greater inequality, rehabilitative efforts based on needs assessment may do little more, if anything, than repair the inequalities that were produced through risk-based sentencing. That is, convicted individuals who are underprivileged face a more severe sentence first, before or combined with access to the resources provided through treatment. Risk assessment and need assessment thus may, in the ‘best’ case, rule each other out, but in the worst case they work together in such a way that social inequalities are reinforced. Basing sentencing on socioeconomic marginality thus raises questions about the effectivity of rehabilitative efforts particularly for high-need offenders who are also at, by definition, high-risk.

An essential additional analytical step is to look at decisions in practice. If risk/needs assessment is indeed an ‘actuarial illusion’ (Hannah-Moffat et al., 2009), because decision-makers override the outcome of risk assessment tools, and if they do so to correct for bias, the discriminatory effects of risk assessment may be absent or less pronounced. However, disparate outcomes may also be exacerbated, when decisions to override or not are based on estimations of deservingness of individuals. Either way, such practices raise questions about the legitimacy of including socioeconomic factors in risk assessment tools. To conclude, there are too many moral and social as well as practical questions that have been glossed over in the decision to allow the assessment of socioeconomic marginality for making sentencing decisions. It is hoped that this analysis will stimulate further research and debate among criminologists and legal scholars, as well as among policymakers and practitioners in the criminal justice field, assessing risk assessment in light of not only its hoped for result, but also in relation to unintended social consequences.