We are in the process of performing a study exploring associations between PCL-R facets and factors and crime-scene behavior in a group of Swedish homicide offenders. After having collected PCL-R results for 72 offenders, we are now ready to perform analyses, in which we will categorize and compare offenders, first, with regard to homicide manner (expressive or instrumental), and, second, with regard to homicide motive (family-related, altercation, criminal conflict, burglary/robbery or sexual).
I have a question regarding choice of statistical test for the analyses. There are multiple non-overlapping independent variables in the two sets of analyses (2 manners, on the one hand, and 5 motives, on the other); whereas, among the chosen dependent variables, each factor is interrelated with its components facets.
It seems to us that, in this situation, an ideal test of statistical probability would first compare groups at the factor level and then, in the event of a significant difference, identify the source of the difference at the facet level by means of a post hoc test.
Is anyone aware of how PCL researchers have handled this situation in the past? I am grateful for any advice anyone has to offer.
The analysis to be used really depends on your specific research questions and hypotheses, which I don't know at this point. You do not seem to have a high enough N to perform too many breakdowns in your sample because these would probably create very small groups. Therefore, I don't think you can look at manner and motivation in interaction. Also, facets are simply a decomposition of the higher-order factors so I am not sure there is a need to look at these solutions separately. Keep in mind that the two factor scores often show moderate correlations.
Providing that the sample size allows for it, you could first look at instrumental vs expressive, then subsequently the 5 motive groups, and the facet scores, using 2 separate MANOVAs. Of course, look for outliers, homoscedasticity and the associations between the facets to make sure it is not too high.
Another option would be to cluster offenders based on their facet scores. Subsequent analyses would then become Chi-Square based.
Thanks, David, for your advice. We've now decided to focus on factors only, and to analyze manner and motive separately. More news as it breaks. Thanks again!
I agree with David, I think the answer depends on your a priori hypothesis about psychopathy. If you think there might be overall differences in manner or motive, then you would look at total score first.
But perhaps you have reason to expect -- from theory or prior evidence -- that differences would be driven more by a particular factor, or facet. Then I would begin the analyses at the factor or facet level, rather than the nested approach.
The nested approach (total or factor, then facet) is appropriate if you are exploring your data and don't have a priori predictions about differences.
Thanks, Michael. We do, in fact, have hyptheses about the factors—- and about the heterogeneity of the phenomenon homicide — and have, therefore, decided not to subject total scores or facets to formal testing. Thanks again.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology