Actuarial Measures Research Articles

Interval Estimation of Actuarial Risk Measures

This article investigates performance of interval estimators of various actuarial risk measures. We consider the following risk measures: proportional hazards transform (PHT), Wang transform (WT), value-at-risk (VaR), and conditional tail expectation (CTE). Confidence intervals for these measures are constructed by applying nonparametric approaches (empirical and bootstrap), the strict parametric approach (based on the maximum likelihood estimators), and robust parametric procedures (based on trimmed means). Using Monte Carlo simulations, we compare the average lengths and proportions of coverage (of the true measure) of the intervals under two data-generating scenarios: “clean” data and “contaminated” data. In the “clean” case, data sets are generated by the following (similar shape) parametric families: exponential, Pareto, and lognormal. Parameters of these distributions are selected so that all three families are equally risky with respect to a fixed risk measure. In the “contaminated” case, the “clean” data sets from these distributions are mixed with a small fraction of unusual observations (outliers). It is found that approximate knowledge of the underlying distribution combined with a sufficiently robust estimator (designed for that distribution) yields intervals with satisfactory performance under both scenarios.

Tracking High-Risk, Violent Offenders: An Examination of the National Flagging System

The present study investigated the effectiveness of the Canadian National Flagging System (NFS), a policy initiative intended to identify offenders who are judged to be suitable candidates for a Dangerous Offender (DO) or a Long-Term Offender (LTO) application. Analyses comparing the profiles of 256 flagged offenders and 97 known high-risk, violent offenders indicated that the flagged offenders generally showed less serious and persistent criminality characteristics than the known high-risk, violent offenders. However, scores on actuarial measures of risk demonstrated that both groups comprised especially high-risk offenders. Furthermore, the violent and/or sexual reconviction rates of the flagged offenders were significantly higher than those reported among the typical Canadian male federal offender population. Judged against our expectations, the base rate of DO/LTO designations among the violent/sexual recidivist flagged offenders was also much higher than the one estimated among the general high-risk, violent offender population in Canada. As a whole, the findings suggested that the NFS was successful in appropriately identifying offenders who pose a risk to the community as well as in subsequently responding to this threat by facilitating the use of the DO/LTO provisions. Recommendations for the development of guidelines to assist criminal justice professionals in screening, monitoring, and processing high-risk, persistent offenders are made.

Psychosexual characteristics of sexual offenders and the relationship to sexual reconviction

The purpose of this study was to determine if the Multiphasic Sex Inventory (MSI) could be used to predict sexual reconviction. The MSI was administered to 119 convicted male sex offenders. Reconviction data were analysed using the receiver operating characteristic (ROC) over 2-, 5- and 10-year follow-up periods. The MSI scales Sexual Obsession and Paraphilia (Atypical Sexual Outlet) obtained good accuracy in predicting sexual reconviction over 2- and 5-year follow-up periods. A confirmatory factor analysis of the MSI scales yielded a four-factor solution: Sexual Deviance, Sexual Desirability, Dysfunctional/Justification, and Normal. The Sexual Deviance factor demonstrated good accuracy in predicting sexual reconviction at 2-year follow-up while the Normal factor was a poor predictor of sexual reconviction. Compared against an actuarial risk assessment measure for sexual offenders, the Sexual Obsession, Sexual/Social Desirability, and Sexual Dysfunction scales, and Sexual Deviance factor made a statistically significant contribution independent of the risk scale in predicting sexual reconviction. It is argued that rather than using the MSI solely as an assessment measure of psychosexual characteristics of sexual offenders in treatment programmes, it can be used to provide additional information as part of an assessment of risk of sexual reconviction.

Factors Influencing Treatment Entry in Sex Offenders Against Children

The study examined the psychosocial characteristics of 99 sex offenders against children referred for treatment to a forensic community sex offender programme. The subjects had high levels of substance misuse problems and personality disorder but low rates of Axis I disorder. Subjects accepted for treatment could not be distinguished from those rejected as unsuitable on the Child Sex Questionnaire (CSQ) (Beckett et al., 1994), apart from the question on whether or not subjects believed it was wrong to have sexual contact with children. Substance misuse, antisocial personality disorder and past criminal history did not determine whether subjects were accepted or rejected. Differences in scores on actuarial measures of risk (STATIC-99) were also not significant. Clinicians' ratings of motivation, level of denial and poor social skills were the key factors determining rejection for treatment. Reconviction rates in the treatment group were low (7%) at five-year follow-up.

Predicting Violence Among Forensic-Correctional Populations

Research on violence prediction during the past 2 decades has evolved appreciably in terms of depicting determinants of violence and developing psychometrically sound actuarial measures to predict the probability of future violent behavior. This article provides a brief synopsis of information on predicting violence gained in the past 2 decades, the methodological advances evolved, and finally significant gaps in violence prediction that still exist for future research endeavors.

Sexual recidivism: A review of static, dynamic and actuarial predictors

The need for accurate risk assessment of sexual offence recidivism has never been greater. It is widely accepted that actuarial risk instruments outperform clinical judgement and the literature has recently witnessed a surge of empirically derived actuarial measures. However, in spite of the increased levels of predictive accuracy, actuarial measures have been criticized as being unrepresentative, lacking specificity, and being heavily reliant on static risk factors without taking into account dynamic risk, psychological emotional states and treatment effects. Rather than offering a critique of the actuarial movement, this paper offers a summary of static and dynamic risk factors associated with sexual offence recidivism as identified from the literature. Implications of incorporating dynamic factors into risk assessments and actuarial measures are discussed.

Refusers, dropouts, and completers: measuring sex offender treatment efficacy.

A sex offender program delivered in a medium-security prison followed 109 treatment completers and 37 noncompleters for 2 years after release. Noncompleters, those who refused treatment or dropped out, had 6 times the rate of sexual and violent reoffending relative to completers. Among those who completed the program, however, positive evaluations of treatment change, such as quality of disclosure and enhanced victim empathy, found in posttreatment assessments did not correlate with recidivism. Furthermore, completers did not differ in their rates of recidivism from pretreatment rates predicted by the Static 99, an actuarial measure of anticipated sexual and violent recidivism. We conclude that the program did not influence propensities for sexual and violent recidivism but rather served as a prolonged screening instrument for sex offenders whose failure to comply with treatment attendance predicted higher rates of recidivism.

Actuarial Risk Assessment and Human Rights: A Commentary

Introduction In recent years, criminal justice professionals have increasingly endorsed actuarial measures of risk as the most reliable predictive instruments for decision making (Ericson and Haggerty 1997; Hannah-Moffat and Shaw 2001). Andrews and Bonta define the actuarial approach as involving criteria for decisions that are validated by (2003: 234). In a recent review of best practices in risk assessment, Bonta (2002) indicates that the first guideline for psychologists when assessing risk is to rely on empirically based actuarial risk instruments. The performance of judgement (i.e., determination of risk based on professional opinion and expertise) compared to actuarial risk assessment in predicting criminal behaviour is by now well documented (Bonta, Law, and Hanson 1998; Grove and Meehl 1996; Grove, Zald, Lebow, Snitz, and Nelson 2000). Some have even argued that failure to conduct actuarial risk assessment or consider its results is irrational, unscientific, unethical, and unprofessional (Grove and Meehl 1996; Quinsey, Harris, Rice, and Cormier 1998). In the criminal justice context, judgement frequently fails to consider criminogenic risk factors (i.e., factors that research has found to be associated with criminality) and too often relies on characteristics not associated with criminality and recidivism, resulting in or unsubstantiated decisions (both false positives and false negatives). Grove and Meehl go so far as to argue that the clinical brain is a poor substitute for an explicit regression equation or actuarial table. Humans simply cannot assign optimal to variables, and they are not consistently applying their own weights (1996: 315). They further suggest that judgement is unduly vulnerable to personal prejudice and bias. The movement toward using actuarial risk assessment has permeated the entire criminal justice system. Criminal justice professionals are increasingly familiar with the long list of high-performing actuarial risk assessment instruments (Bonta 2002) with impressive acronyms, such as PCL-R, LSI-R, VRAG, HRC-20, SONAR, RRASOR, and STATIC-99. A few provincial jurisdictions require probation officers to include actuarial risk assessment results in pre-sentence reports (Cole and Angus 2003). Judges frequently rely on actuarial risk assessments when handing down their sentences, and a high level of risk invariably results in harsher dispositions (Zinger and Forth 1998). Correctional authorities routinely depend on actuarial risk assessment to make decisions on security classification, institutional placement, programming, and conditional release (e.g., temporary absences, work releases). Parole boards typically require such assessments before rendering decisions with respect to parole or other forms of conditional release. For many psychologists working in criminal justice, and a few psychiatrists, actuarial risk instruments are a part of the daily work routine and preoccupations (Bonta 2002). Academic journals are filled with articles on the validity of specialized scales that attempt to predict various forms of recidivism (general, violent, sexual recidivism). More telling, perhaps, is the literature comparing the performance of different instruments. The debates surrounding scale performance are frequently heated (Gendreau, Goggin, and Smith 2002; Hemphill and Hare 2004). Those who have developed instruments that outperform their competitors also appear to feel a certain sense of pride and accomplishment (Pratt 2001). Moreover, developing an instrument may result in considerable profit and prestige and, in some rare instances, even celebrity (see The Corporation [Abbott and Achbar 2003], featuring Dr Hare's PCL-R [1991]). In addition to substantial royalties, some instruments may generate further revenues by requiring mandatory specialized training in their use, application, and interpretation (e. …

Personality characteristics associated with reconviction in sexual and violent offenders

The purpose of this study was to examine personality characteristics associated with reconviction and to explore the accuracy of the Special Hospitals Assessment of Personality and Socialization (SHAPS) in predicting reconviction. The SHAPS was administered to a group of sexual and violent offenders (n = 121) who were followed up over 2, 5 and 10 year periods. Reconviction data were analysed using the Receiver Operating Characteristic (ROC). Of the 10 SHAPS scales, the Impulsivity scale produced the most promising results and consistently obtained the highest accuracy in predicting violent and general reconviction. The predictive accuracy of the SHAPS scales was compared against three actuarial risk measures, Static-99, Risk Matrix-2000 Sexual and Violent scales. The strength of correlations and accuracy of the SHAPS Impulsivity scale in predicting violent and general reconviction is comparable to that of the Risk Matrix 2000/Violent actuarial risk measure. However, the SHAPS scales added little to the actuarial measures in a regression equation. The actuarial scales were better at predicting sexual reconviction compared with the SHAPS scales. The implications of assessment and treatment using the SHAPS are discussed.

Limitations in actuarial risk assessment of sexual offenders: a methodological note

The assessment of risk of recidivism in sexual offenders is fundamental to clinical practice. It is widely accepted that, compared with actuarial measures of risk, unaided clinical judgment has generally been found to be of low reliability. Consequently, the literature has shown a surge in actuarial measures. However, a major difficulty in assessing risk in sex offenders is the low base rate, leading to an increased likelihood of making a false positive predictive error. To overcome this, risk assessment studies are increasingly using the receiver operating characteristic (ROC), which displays the relationship between level of risk and decision choice. This note summarises the methodological issues in measuring predictive accuracy in assessing risk of re‐offending in sexual offenders, and identifies from the literature both static and dynamic risk factors associated with sexual offence recidivism.

Survival of blacks and whites after a cancer diagnosis.

In recent years a theory that cancer biology is different in blacks and whites has gained prominence in reaction to epidemiologic observations that blacks have poorer survival than whites, even when diagnosed with cancer of similar severity. Yet, few studies have evaluated whether lower-quality treatment and shorter overall life expectancy due to a greater burden of other illnesses may explain the survival discrepancy. To estimate the magnitude of overall and cancer-specific survival differences between blacks and whites who receive comparable treatment for similar-stage cancer. We searched MEDLINE for English-language articles published from 1966 to January 2002 that reported on overall survival for black and white patients treated similarly for cancer. The abstracts or titles for 891 citations were independently examined by 2 authors. The full text was retrieved if the abstract mentioned both black and white patients, made some comment regarding either similarity of treatment received or presented an analysis based on the treatment received, and commented on survival. Studies were included if they included data for at least 10 black and 10 white patients; specified the cohort ascertainment method and what measures were undertaken to minimize loss to follow-up; summarized survival of both blacks and whites using actuarial measures; presented outcomes within stage, adjusted for stage, or based on cohorts with balanced stage distributions; and specified that blacks and whites in the study received similar treatment. We identified 89 unique cohorts in 54 articles that met our inclusion criteria. Overall survival rates and hazard ratios (HRs) for death for blacks relative to whites were calculated. These were subsequently adjusted for rates of death due to causes other than the cancer under study to determine cancer-specific survival and cancer-specific HRs. Results represent 189 877 white and 32 004 black patients with 14 different cancers. Compared with whites, blacks had an overall excess risk of death (HR, 1.16; 95% confidence interval [CI], 1.12-1.20). After correction for deaths due to other causes, the cancer-specific HR was 1.07 (95% CI, 1.02-1.13). Of the 14 cancers, blacks were at a significantly higher risk of cancer-specific death only for cancer of the breast, uterus, or bladder. Only modest cancer-specific survival differences are evident for blacks and whites treated comparably for similar-stage cancer. Therefore, differences in cancer biology between racial groups are unlikely to be responsible for a substantial portion of the survival discrepancy. Differences in treatment, stage at presentation, and mortality from other diseases should represent the primary targets of research and interventions designed to reduce disparities in cancer outcomes.

Preaching to the Choir: A Comparison of the Use of Integrated Data Sets in Criminology Journals in Australia, England and the United States

Empirical analyses of crime increasingly rely on integrated data. This paper considers advantages and limitations of integrated data sets, comparative uses of such data in The Australian and New Zealand Journal of Criminology, British Journal of Criminology and Criminology, as well as speculations based on these findings. Integrated data analysis has amplified methodological issues. The reliability of secondary data analysis is both supported and challenged. Relevant questions include how crime is associated with actuarial measures and which measures are most reliable. The more difficult question is whether actuarial measures are valid indicators. One extreme empirical orientation, positivism, relies on empirical, often quantitative, data, and scientific method to derive conclusions and guide policy (Young, 1992a). The other extreme relies on discourse and persuasion. Implications of relying on or rejecting actuarialism are discussed in the context of integrated data research published in the above-mentioned criminology journals.

Preaching to the Choir: A Comparison of the Use of Integrated Data Sets in Criminology Journals in Australia, England and the United States

Empirical analyses of crime increasingly rely on integrated data. This paper considers advantages and limitations of integrated data sets, comparative uses of such data in The Australian and New Zealand Journal of Criminology, British Journal of Criminology and Criminology, as well as speculations based on these findings. Integrated data analysis has amplified methodological issues. The reliability of secondary data analysis is both supported and challenged. Relevant questions include how crime is associated with actuarial measures and which measures are most reliable. The more difficult question is whether actuarial measures are valid indicators. One extreme empirical orientation, positivism, relies on empirical, often quantitative, data, and scientific method to derive conclusions and guide policy (Young, 1992a). The other extreme relies on discourse and persuasion. Implications of relying on or rejecting actuarialism are discussed in the context of integrated data research published in the above-mentioned criminology journals.

Valuing the strategic option to sell life insurance business: Theory and evidence

We present a simple put option pricing procedure within an asset–liability valuation model that can be used to estimate the incentives facing stock-based life insurance firms to voluntarily sell their businesses under various operating and regulatory conditions. Estimates are derived for samples of 11 sold firms and 24 continuing Australian life insurance companies over a period of industry consolidation. The put option values interact with other actuarial and accounting components of the fair value of these life insurance firms and are used to assess the effectiveness of accounting and actuarial measures of capital, under static or dynamic based solvency testing models.

Anger and Prediction of Violent and Nonviolent Offenders' Recidivism

This study investigates the usefulness of assessing anger for the prediction of violent and nonviolent recidivism. Four risk/need measures and four anger inventories were administered to 252 Canadian federally incarcerated male offenders. The association between anger and recidivism was investigated by (a) comparing groups of participants classified to different levels of risks for recidivism according to their scores on actuarial measures of recidivism and their scores on measures of anger, (b) canonical correlation between the offenders' scores on measures of risk for recidivism and the anger inventories, and (c) correlating the clinical variables of age at admission, number of past offenses, and number of violent offences with offenders' scores on measures of anger. The results indicate that all comparisons are not significant. Results of this study shed doubts on the usefulness of assessing anger when predicting violent and nonviolent recidivism.

Predicting Prison Misconducts

A meta-analysis was conducted on 39 studies that generated 695 correlations with prison misconducts. Predictors of prison misconducts were grouped into 16 domains as follows: (a) personal characteristics ( n = 9), (b) situational factors ( n = 3), and (c) actuarial measures of antisocial personality and risk ( n = 4). Personal and situational variables were similar in their ability to predict prison misconduct. Within these two categories, antisocial attitudes and behavior (e.g., companions, prison adjustment), criminal history, and institutional factors were the strongest predictors. Among actuarial measures, an interview-based risk protocol produced the highest correlations with prison misconducts. The prediction of violent misconducts was associated with greater effect sizes than nonviolent misconducts. Despite the limitations of the database, several recommendations for assessing prison misconducts appear warranted.

A META‐ANALYSIS OF THE PREDICTORS OF ADULT OFFENDER RECIDIVISM: WHAT WORKS!*

Meta‐analytic techniques were used to determine which predictor domains and actuarial assessment instruments were the best predictors of adult offender recidivism. One hundred and thirty‐one studies produced 1,141 correlations with recidivism. The strongest predictor domains were criminogenic needs, criminal history/history of antisocial behavior, social achievement, age/gender/race, and family factors. Less robust predictors included intellectual functioning, personal distress factors, and socioeconomic status in the family of origin. Dynamic predictor domains performed at least as well as the static domains. The LSI‐R was identified as the most useful actuarial measure. Recommendations for developing sound assessment practices in corrections are provided.