The present study was reviewed and approved by the Committee of the Faculty of Psychology of the Complutense University of Madrid. Written informed consent was obtained from all participants. All participants were informed of the purpose of the research, the estimated duration of the assessment, and the procedures that were followed for the completion of the research.

Two-hundred and seventy-six male offenders who were incarcerated in different prisons in the Community of Madrid (Spain) participated at the present study. The average age of the participants was 36.4 years (SD=9.7, ranging from 19 to 66 years). Of them, 91.3% were Spanish, and 8.7% were of other nationalities. Table 1 displays the demographic characteristics of the offenders. Current sentence length ranged from 8 to 288 months (M=7.39 years, SD=6.47). Seventy-five percent of the participants presented violent offenses (i.e., an offense against a person such as murder, rape, and assault), and the nonviolent offenders were convicted of nonviolent crimes such as property offenses (i.e., robbery without violence, breaking and entering, and theft). See Table 2 for criminal characteristics of the offenders.

The Psychopathy Checklist–Revised (PCL-R; Hare, 1991, 2003). The PCL-R consists of 20 items and two factors assessing psychopathy in clinical and forensic populations. Several studies have supported the psychometric properties of the PCL-R as well as its use as a predictor of violence (Mokros, Vohs, & Habermeyer, 2014; Olver, Lewis, & Wong, 2013; Piquero et al., 2012; Walters, 2012). The Spanish standardization of the PCL-R (Moltó, Poy, & Torrubia, 2010) was used in the present study.

The Violence Risk Appraisal Guide (VRAG; Harris, Rice, & Quinsey, 1993). The VRAG is a measure developed for assessing the risk of violent recidivism in penitentiary and forensic populations. The instrument displays the probability that an offender will commit another violent crime. Higher scores indicate a greater risk of reoffending violently. The instrument consists of a list of 12 items: 1) lived with both biological parents till age 16 (except for death of parent); 2) elementary school maladjustment; 3) history of alcohol problems; 4) marital status (at the time of or prior to index offense); 5) criminal history score for nonviolent offenses prior to the index offense; 6) failure on prior conditional release (includes parole or probation violation or revocation, failure to comply, bail violation, and any new arrest while on conditional release); 7) age at index offense; 8) victim injury for index offense; 9) any female victim for index offense; 10) meets the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria for any personality disorder; 11) meets DSM criteria for schizophrenia; and 12) Psychopathy Checklist-Revised score. The Spanish version of the VRAG (Ballesteros, Graña, & Andreu, 2006) was used at the present study.

The Self-Appraisal Questionnaire (SAQ; Loza, 2005). The SAQ is a self-report instrument designed to predict violent and nonviolent recidivism and assess treatment needs in correctional and forensic populations. The questionnaire is composed of 72 true-false items that comprise eight subscales: 1) Criminal Tendencies; 2) Antisocial Personality; 3) Conduct Problems; 4) Criminal History; 5) Alcohol and Drug Use; 6) Antisocial Associates; 7) Anger; and 8) Validity. It is important to note the Validity subscale is included in the other SAQ subscales, and the specific items on this subscale allow one to validate the offender's accuracy when responding to the SAQ items. In particular, these items ask about prior arrests or convictions. Complementarily to the SAQ total score, the score for each subscale can be calculated to assess potential treatment targets (Loza, 2005).

The SAQ measures the predictive variables found in the literature to be related to violent and nonviolent recidivism. The SAQ total score yields an estimation of the offender's level of risk and determines the offender's probability of recidivism, with higher scores indicating a greater level of risk and recidivism. According to Loza and Loza-Fanous (2003), the probability of recidivism can be estimated from the total score as low (SAQ total score=2 to 19), medium (20 to 30), and high (31 to 58). The Spanish translation of the SAQ (available in Andreu, 2010) was used in the present study.

Participation in the present study was voluntary and the following inclusion criteria were used: a) speaking Spanish; b) having a definitive judicial verdict and sentence; and c) having given written informed consent to participate in the research. All participants were informed that the information provided would be used for research purposes, and that all their data would remain confidential. They were individually interviewed and assessed by forensic psychology students from the Master's Degree in Clinical, Legal and Forensic Psychology (Complutense University of Madrid), who were trained in the administration of the risk assessment protocols used in this study. Offender's criminal history was obtained from penitentiary files.

Internal consistency of the SAQ was examined with Cronbach's alpha. Pearson correlations were used to analyze the concurrent validity of the SAQ with related variables of violence risk. Group differences comparing recidivist groups of offenders were analyzed using Student's t-test, determining the effect size with the partial eta squared coefficient. To examine the SAQ's predictive efficiency for recidivism, the ROC curve was taken into account. All analyses were performed with the SPSS software (v. 22.0).

Table 3 shows the internal consistency for the SAQ total score and subscales. The alpha coefficient for the SAQ total score was .92, and alpha coefficients for the subscales ranged from .35 (antisocial associates subscale) to .83 (conduct problems subscale).

Concurrent validity was analyzed by correlating SAQ scores with two instruments that show adequate psychometric properties for violence risk assessment. Pearson correlation coefficients between all measures are shown in Table 4. The SAQ total score showed moderate correlations with the PCL-R total score (r=.47, p<.01) and higher correlations with the VRAG total score (r=.65, p<.01). Correlations between the SAQ subscales and the PCL-R total score ranged from .20 to .47 and correlations with the VRAG total score ranged from .34 to .65. Analysis of the correlations between the SAQ subscales and factors 1 and 2 from the PCL-R showed that antisocial associates and anger did not show any significant correlations with PCL-R Factor 1. The rest of the SAQ subscales displayed higher correlations with PCL-R Factor 2, ranging from .36 (anger) to .53 (criminal history), in comparison with PCL-R Factor 1, whose correlations ranged from .13 (criminal tendencies and conduct problems) to .22 (antisocial personality). Lastly, the SAQ total score had a higher correlation with PCL-R Factor 2 (r=.66) than with Factor 1 (r=.13).

In order to examine the predictive validity of the SAQ, the following criterion was used as a measure of recidivism: a second or subsequent entry in prison by the same person for committing a violent crime in the community, which may have been committed after serving the sentence or while enjoying a temporary leave from prison. A violent crime is defined as a felony involving acts reasonably regarded as likely to harm other people, clearly threatening behaviors, sexual aggressions, and the destruction of objects (Andrés-Pueyo & Echeburúa, 2010). The sample was divided into two groups: violent recidivists (individuals who committed a new crime qualified as violent during the temporal period ranging from 2007 and 2011) and nonviolent recidivists (primary offenders or individuals who committed a new offense considered nonviolent) during the same temporal interval in which retrospective recidivism of the offenders was estimated.

Table 5 shows the comparison of both groups in the SAQ: the means for the violent recidivist group were higher in all subscales and SAQ total score than the means of the nonviolent recidivist group, and the differences were statistically significant at p<.001. The highest value for the partial square eta coefficient corresponded to alcohol and drug abuse (.11), criminal history (.09), and SAQ total score (.09).

The next step in the examination of the validity of the SAQ was though investigation of the SAQ's predictive efficiency. Using the receiving operator characteristic (ROC) curve, the diagnostic efficacy of the SAQ was described in terms of sensitivity (probability that the measure correctly classified violent recidivist participants) and specificity (probability that the measure correctly classified the nonviolent recidivist participants). Values for sensitivity and for false-positive rates (1 – specificity) were plotted, and the area under the curve (AUC) represented the probability that the measure correctly classified participants as positive or negative with respect to recidivism. This area theoretically ranges from 0.5 (inaccuracy in discriminating) to 1.0 (perfect accuracy). The following traditional point system is a rough guide for classifying accuracy: (a) .90-1.0=excellent, (b) .80-.90=good, (c) .70-.80=fair, (d) .60-.70=poor, and (e) .50-.60=failure.

The ROC analysis carried out on the SAQ total score revealed an AUC of .80 (95% CI [0.73, 0.87]), showing acceptable accuracy discriminating between violent and nonviolent recidivist. Thus, there would be an 80% probability that the diagnosis of a violent recidivist was correct. Figure 1 presents the ROC curve for SAQ total score and violent recidivism.

Figure 1.

ROC Curve for SAQ Total Score and Violent Recidivism.

(0.09MB).

The best cut-off point that discriminated between violent and nonviolent recidivists was 32 (see Table 6), with a sensitivity of 80% and a 1-specificity of 30%. Thus, in eighty percent of the cases, a score of 32 correctly classified 80% percent of the violent offenders and 70% of the nonviolent recidivist group.

Figure 2 presents the ROC curves comparing the predictive efficiency of the SAQ, the PCL-R, and the VRAG for violent recidivism. The AUC of the PCL-R (AUC=.63) was lower than that of the SAQ (AUC=.80), whereas the VRAG obtained the highest AUC (.87).

Figure 2.

ROC Curves Comparison for SAQ, PCL-R, VRAG and vVolent Recidivism.

(0.11MB).