The Level of Service Inventory (LSI) is a risk assessment instrument that was designed to identify adult offenders’ likelihood of reoffending. As such, it is one of a number of offender risk assessment tools that are now used by a wide range of correctional, forensic, and mental health professionals who are given the difficult task of predicting offender recidivism. Various versions of the LSI are now used extensively in Canada, the United States, and various countries in Europe, Asia, and Oceana, in large part because they have demonstrated their capacity to predict offender recidivism among a wide range of offenders and they shed light on areas that correctional officials should target with the offender to reduce his or her likelihood of recidivating. Consequently, the LSI is described as a risk/need scale in that it assesses both the overall risk that an offender presents in terms of likelihood to recidivate and the criminogenic needs that might be addressed to reduce that risk. This article describes the origins and history of the LSI, its theoretical and empirical basis, its design and content, and its research support.
Origins and History
The original LSI was developed in the early 1980s by Donald Andrews, in Ottawa, Canada, with the cooperation of the Ontario Ministry of Correctional Services. Andrews was originally interested in assisting the Ministry to establish supervision standards for probation officers. He was intrigued by the work of Sally Rogers, a Ministry researcher, who came up with a simplified instrument that effectively identified offenders who recidivated at both above average and below average rates of recidivism. Rogers’s 6-item scale asked yes–no questions about offenders’ demographics and criminal history (i.e., male, under 24 years of age, have criminal record, have delinquent associates, aimless use of leisure time, have family on social assistance). But Andrews was aware of more complex tools that had been developed in the United States, including the Salient Factor Score. He also believed that an instrument that was grounded in a sound theory of criminal behavior could do a better job of prediction. This led to his first effort, known originally as the Level of Supervision Inventory, an offender risk assessment tool that was guided by various theories about criminal behavior and empirical evidence.
The Theoretical and Empirical Foundations
Andrews was strongly influenced by the differential association theory of Edwin Sutherland and Donald Cressey and its more behavioral reformulations and by Albert Bandura’s more general social learning theory. Differential association purported that criminal behavior was a learned behavior and that such learning took place primarily among intimate personal groups. Donald Akers then applied more Skinnerian terminology emphasizing the rewards and punishments that occurred as consequences of antisocial behavior, while Bandura introduced the concept of modeling (i.e., behavior, including antisocial behavior, can be learned by simply observing others and is strengthened when the model is rewarded).
This led to Andrews considering not only delinquent and criminal history as factors in a risk assessment tool but also an offender’s history of support and punishment for both prosocial and antisocial behavior. He envisioned the brain as an unconscious computer, which routinely tallies all of the rewards and punishments (i.e., positive and negative consequences) for different kinds of behavior. However, the nature of the recent or current environment in which the offender is living is most important as it constantly provides a wide variety of rewards and punishment to a wide range of behavior. He reasoned that these rewards and punishments come in the form of material gain or loss, social or interpersonal support or criticism, and intrapersonal thoughts (congratulatory or critical) about the offender’s behavior after the behavior has been performed. Andrews formalized this conceptualization as the Personal, Interpersonal Community–Reinforcement (PIC-R) perspective of criminal behavior. Years later, based on their appreciation of personality and attitudes, Andrews and James Bonta expanded the PIC-R concept by including it in a more integrated theoretical model of criminal behavior called a general personality and cognitive social learning perspective of human behavior including criminal behavior, which helped to frame the LSI and its successors.
Another important development by Andrews and Bonta, along with Robert Hoge, that spoke directly to the development and evolution of the LSI was the empirically derived principles of risk, need, and responsivity. The risk principle directs correctional workers to focus their supervision and attention on moderate- and high-risk offenders. The need principle directs correctional workers to target the identified criminogenic needs of an offender in his or her rehabilitation process. The responsivity principle directs workers to adjust their style of interaction with offenders according to the offenders’ cognitive, personality, motivational, and demographic characteristics so as to optimize supervision and intervention efforts. GPSLC and risk, need, and responsivity also comprise the backbone of what Andrews and Bonta have referred to as the psychology of criminal conduct. The development of these concepts occurred hand in glove with the LSI and its subsequent revisions. Therefore, a good understanding of the PIC-R model, the risk-need-responsivity principles, and the general personality and cognitive social learning perspective is needed in order to appreciate the LSI, its content, and its evolution over time.
It is important to note that PIC-R and the general personality and cognitive social learning were strongly supported by empirical research, most of which was psychological, sociological, and criminological in nature. Some of this work harkened back to classical, cross-sectional comparison of delinquents and nondelinquents conducted by Sheldon and Eleanor Glueck in the 1940s and 1950s, in which the two groups were found to be different on a wide array of dimensions, including attitude, disposition or personality, school behavior and performance, relationship with and supervision by parents, and peer relations. In 1969, a second major empirical study by Travis Hirschi revealed a similar list of characteristics that differentiated delinquents from nondelinquents and guided Andrews in his initial effort to develop a risk assessment tool. Numerous similar studies followed, allowing Andrews and others, including Paul Gendreau, to conduct meta-analyses of the predictors of criminal behavior and recidivism that continued to guide subsequent versions of the LSI through the 1990s and 2000s.
Rationale for Development
With the rise of offenders requiring community supervision and growing caseloads for probation officers, it had become clear that some means were required to direct the increasingly scarce supervision resources to offenders who were at highest risk of recidivating. Moreover, evidence was emerging that the kind of service and the specific focus of intervention was important to the success of offenders in the community. In other words, to which offenders should probation officers devote their attention (risk principle) and on what issues should that attention focus (need principle)?
Until the early 1980s, those questions were addressed neither by theoretically sound or empirically driven practices. Instead, offender risk assessments were of two types: either they were based on subjective professional judgment, which, it was being learned, were little better than chance and were described as the first generation of risk assessment by Bonta or they were based on historical or demographic characteristics that were not subject to change and were described as second generation risk assessments. Although second generation risk assessments did offer assistance in terms of suggesting the amount of supervision (i.e., more supervision for high-risk offenders), such as the Salient Factor Score in the United States or Joan Nuffield’s Statistical Information on Recidivism Scale in Canada, they did not offer any direction to the correctional worker in terms of the content of supervision simply because they consisted only of nonchangeable, static risk factors. Meanwhile, traditional psychological offender assessments were based on personality characteristics that had little or nothing to do with criminal behavior. The LSI was designed to rectify these shortcomings by offering a tool with practical value to correctional workers and their agencies.
The LSI is described as a third generation or risk/need assessment tool since it includes both static risk items and dynamic (i.e., changeable) criminogenic need items. There are two particular advantages to including dynamic risk items in an offender recidivism prediction. First, as the situation and conditions of the offender change over time, an accurate reassessment will reflect these changes in risk that are occurring either naturally or by means of some kind of treatment or intervention. Secondly, dynamic criminogenic needs are, by definition, the characteristics of an individual, which, when changed, translate to changes in offender recidivism, thus giving the probation officer or other correctional worker or service provider direction as to how to facilitate the rehabilitation of the offender.
Design and Content
Andrews’s goal was to develop a wide array of predictor items that could capture the diversity of risk factors that contribute to offender recidivism. To achieve that goal, he conducted a thorough review of offender documents on file and consulted with collateral individuals who could provide insight about offenders. Andrews then realized that more detailed information than what was routinely available in offender files was required so he developed an interview guide to unearth some of the more nuanced aspects of offender risk. He also worked with data sets of probationers and their outcomes that were provided by the Ontario Ministry of Correctional Services. Items were added and deleted in a series of revisions in an effort to capture not only the best combination of predictors but also ones that could be used to guide probation officers’ supervision of their clients. Probation officers and their managers were also consulted in the choice of items and the design of the instrument. The result was the Level of Supervision Inventory VI.
The LSI-VI consisted of 58 items that were grouped into 11 sections. These sections and the number of items in each section (in brackets) included the following: Criminal History (10), Education/Employment (10), Financial (2), Family/Marital (3), Accommodation (3), Leisure/ Recreation (2), Companions (5), Alcohol/Drug Problems (9), Emotional/Personal (5), Probation/Parole Conditions (4), and Attitudes/Orientation (4). It is readily apparent that some domains included many more items than others. As each item counted as a single point, varying the number of items in a domain allowed Andrews to weight the different domains in accordance with their perceived importance as determined by the research at the time. Specifically, criminal history, education and employment, and alcohol and drug problems were perceived as being most strongly related to criminal recidivism, while financial issues, leisure and recreation activities, and accommodation were relegated to minor risk domains. The total score was the sum of all items endorsed and was used to determine a level of supervision for the probationer.
But the LSI-VI also included a scoring quirk. Realizing that many of the criminogenic need factors existed on a continuum, Andrews established a 0–3 rating scale to describe the degree to which a positive attribute, or strength, is present. For example, a maximum rating of 3 on the Employment item, Participation/Performance, occurs when the offender expresses a strong interest in job, takes pride in abilities/performance, has reliable attendance, is willing to work overtime, and wants to stay in the same line of work. A minimum rating of 0 occurs when the offender says he or she hates work, can’t perform well, has unreliable attendance, and is often late. This finer level of detailed scoring can be used to track minor adjustments, improvement or deterioration, and can assist with the case management of the offender. However, these items are reduced to a binary (0 or 1 = 1 and 2 or 3 = 0) for the purpose of calculating the total risk/need score. This format has been retained in all subsequent versions of the LSI.
The preceding description of the Employment item, Participation/Performance, illustrates three important aspects of the LSI items. First, it taps into dynamic aspects of offender risk that exist along a continuum and are subject to change, sometimes rather quickly. Second, items are true to social learning theory and the PIC-R principles in that they are based on the eligibility for different kinds of rewards and punishments for prosocial and antisocial behavior in the offender’s environment. (Note that the next 2 items in the LSI-VI capture the nature of the offender’s relationship with peers and with authority figures in the workplace.) Third, the LSI-VI considers both the presence of risk items and strengths (i.e., protective items) with the absence of a strength being scored as 1. Critics of the LSI often fail to appreciate this means of considering offender strengths.
Another feature introduced in the LSI-VI (and maintained through all subsequent versions of the LSI) included an opportunity for the assessor, typically a probation officer, to override the score-derived level of supervision. A text box was included for assessors to note both positive and negative circumstances that may not have been given sufficient attention in the established LSI-VI items. Pending charges, extreme violence, or particularly notorious actions might be used to elevate the degree of supervision, while positive circumstances might include prosocial supportive factors that would enable the assessor to reduce the level of supervision suggested by the LSI-VI score.
Andrews then wrestled with the number of risk levels into which the raw score could be collapsed, both for statistical analysis and for correctional workers to describe the degree of risk presented by an offender. An initial scheme created four levels of risk: 0–7, 8–11, 12–23, and 24–58. But this allocation was quickly reduced to three risk groups to correspond with the three traditional levels of supervision in community corrections (i.e., minimum, medium, and maximum). This was achieved by combining the two highest risk groups in part because only about 4% of the probationers in the early research scored over 24 in spite of the fact that this was a 58-point scale. Subsequent versions of the LSI have used four and even five levels of risk.
Initial Research
Research on the LSI looked promising right from the start. Beginning in the early 1980s, two important reports to the Ontario government summarized an impressive list of results. To begin with, agreement between raters was considered high and was generally consistent with probation officer judgments pertaining to the required supervision of offenders at the outset of their probation and later, midway through the supervision process, both in the order of 90% agreement. Moreover, LSI scores were consistent with the actual number of supervision hours as documented in case records. Low LSI scores corresponded with probationers who were granted an early case closure, while high LSI scores corresponded with in-program recidivism (i.e., charges and reconvictions), multiple reconvictions, reincarceration, and self-reported criminal behavior. The LSI also predicted post-probation recidivism over a 3-year period, with increased accuracy as follow-up time increased. Working in a detention facility, Bonta found that the LSI predicted failure among prisoners who were referred to a halfway house.
Moreover, a number of LSI sections were related to other self-reported measures of related constructs. For example, criminal history was related to self-reported crime, education/employment was related to inadequacy–immaturity, alcohol/drugs was related to self-reported drug offenses, and personal/emotional was related to sensation seeking and negatively to self-control.
The second review not only replicated many of the original findings but also expanded on them. LSI subsection scores correlated with a host of well-established paper-and-pencil psychometric measures of related constructs. Probation intake LSIs were strongly correlated with in-program recidivism. LSI scores correlated with both official and self-reported (i.e., undetected) crime. Importantly, probationer progress signaled changes in their outcome in that retest LSIs were more predictive of recidivism than intake LSIs, and changes on an LSI progress assessment were associated with corresponding changes in the chances of recidivism. The allocation of more intensive services to high-risk offenders was rewarded with disproportionately lower recidivism rates than when intensive services were allocated to low-risk offenders.
In Australia, when LSIs were conducted in conjunction, but not submitted, with presentence reports, the presentence LSI was associated with the disposition of the court: Those who received fines or acquittals received the lowest scores, followed by short-term probation, long-term probation, and finally a prison term. These results led to an obvious question: Are less serious offenders at lower risk or are offenders who are lower risk more likely to get less severe sentences? This and other findings have generated continued discussions about conducting an LSI prior to sentencing, or even conviction, and the role of the LSI in the sentencing process.
Final Thoughts
By the end of the 1980s, the LSI had established a solid research-based, respected position in Ontario and garnered a selective following beyond. Its reputation was based on the following:
- It provided a standardized record to guide the discretionary controls and services offered by the correctional agency.
- It corresponded with actual levels of supervision that were being applied at the time and correlated with correctional outcomes to an acceptable degree of accuracy.
- It demonstrated what became known as the risk principle by showing that augmented services were wasted on low-risk offenders but reduced recidivism when used on high-risk offenders.
- The LSI captured numerous empirically demonstrated factors related to criminal behavior.
- Its demonstration that assessments of change corresponded to changes in outcome encouraged more efficient and effective allocation of correctional resources.
- Finally, these studies offered robust evidence for the predictive validity of the LSI across a wide range of circumstances, including various types of offenders, several probation offices, various measures of offender outcome, and in different criminal justice contexts, including court, probation, and prison.
References:
- Andrews, D. A. (1982). The Level of Supervision Inventory (LSI): The first follow-up. Toronto, Canada: Ontario Ministry of Correctional Services.
- Andrews, D. A., & Bonta, J. (2010). The psychology of criminal conduct (5th ). New Providence, NJ: LexisNexis.
- Andrews, D. A., Bonta, J., & Hoge, R. D. (1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17, 19–52.
- Andrews, D. A., & Robinson, D. (1984). The Level of Supervision Inventory (LSI): Second report. Report to Research Services. Toronto, Canada: Ontario Ministry of Correctional Services.
- Bandura, A. (1977). Social learning theory. New York, NY: General Learning Press.
- Glueck, S., & Glueck, E. T. (1950). Unraveling juvenile delinquency. Cambridge, MA: Harvard University Press.
- Hirschi, T. (1969). Causes of delinquency. Berkley: University of California Press.
- Sutherland, E. H., & Cressey, D. R. (1970). Principles of criminology (6th ed.). New York, NY: Lippincott.