The reliability check
As was conveyed in our main coding description above, the degree of instrumentality or reactivity associated with a homicide mainly considered (a) instrumental gain, (b) impulsivity, and (c) level of antecedent affective arousal. We assumed that these dimensions would generally be closely interrelated in considering homicidal violence. For example, we predicted that a clear instrumental gain would generally be associated with low impulsivity and low affective arousal. However, it was important to ex- plore empirically how these three main components co-occurred and how each related to both the instrumentality ratings and the PCL–R scores. Therefore, these three dimensions were coded for the entire sample of the homicides (and dual coded for a reliability check in 19 [15.2%] cases) to allow a careful delineation of their interrelationships and relationships with instrumental violence and psychopathy. Trained raters coded for the pres- ence or absence of an instrumental gain (evidence or no evidence), impul- sivity (not, somewhat, or highly impulsive; Hare, 1991), and affect arousal (low, moderate, or high amount of emotional arousal). (Interested readers may contact the authors for more detailed information on how the three dimensions were coded.)
Results
Preliminary Analyses
PCL–R scores and interrater reliability. For the entire sample, the mean PCL–R total score was 22.27 (SD ” 8.81; range ” 1–37). Using the diagnostic cut-off score of !30, offenders were classified either as psychopaths or nonpsychopaths. Ninety-one (72.8%) offenders scored below the cut-off and were classified as nonpsychopaths, whereas 34 (27.2%) offenders scored within the psychopathic range.
A preliminary interrater reliability check was conducted on the PCL–R scores, using 21 (16.8%) randomly selected case files for dual coding. Intraclass correlation coefficients (ICC) were exam- ined to determine the level of inter-rater reliability for continuous scores. Interrater reliability was high/acceptable for PCL–R total, Factor 1, and Factor 2 scores (ICCs ” .92, .81, and .95, respec- tively; ps # .001). Further, there was no mean difference between the two sets of scores (M ” 24.95, SD ” 7.91, and M ” 25.81, SD ” 6.91, for Rater 1 and Rater 2, respectively), t(40) ” .37, p ! .05. Similarly, computing Cohen’s kappa revealed an acceptable level of agreement between Raters 1 and 2 for classifying the offenders as psychopaths or nonpsychopaths, kappa ” .79, p # .001 (common guidelines for acceptable kappa scores are #.40 ” poor; .40–.59 ” fair; .60–.74 ” good; and !.75 ” excellent; e.g., Cicchetti & Sparrow, 1981; Fleiss, 1981).
To examine the potential problem of circularity or criterion contamination, a second interrater reliability check was conducted on an additional 33 (26.4%) randomly selected files in which the
439PSYCHOPATHY AND HOMICIDE
Sylviane Houssais
raters were kept blind to the description of the offenders’ current homicide offense (the description of the offense was removed in advance of coding). Similar to the initial reliability check, inter- rater reliability was high/acceptable for PCL–R Total, Factor 1, and Factor 2 (.97, .95, and .94, respectively; ps # .001). This analysis established that the PCL–R ratings were valid and argues against the possibility of circularity in the ratings. Interrater reliability for homicide coding. An interrater reli-
ability check was conducted on the variables coded pertaining to the characteristics of the homicides. A second well-trained rater who followed the same coding guidelines as the first rater coded the homicide variables (the second rater had been trained over a 2-day period and was kept blind to the PCL–R data for each offender). The reliability check (using 21 randomly selected files) indicated that the coded scores were highly reliable. Specifically, reliability was high/acceptable for type of homicide, K(21) ” .81, p # .001, and specific type of instrumental violence, K(21) ” .87, p # .001.1 An interrater reliability check also was conducted on the three dimensions of the homicide offenses, using 19 randomly selected case files. ICCs were used to determine the level of interrater reliability for these scores. Inter-rater reliability was high/acceptable for gain, impulsivity, and affect (ICCs ” .90, .95, and .88, respectively; ps # .001).
Descriptive Statistics
Age. The offenders’ mean age at the time of data collection was 41.8 years (SD ” 10.5; range ” 18–67 years). The mean age at the time they committed the current homicide was 30.0 years (SD ” 9.5; range ” 14–55 years). The age at which the offender committed the current homicide did not differ significantly be- tween psychopaths and nonpsychopaths (p ! .05). Characteristics of the victims. The victims’ mean age at time
