Subjects were recruited through workshops where participants, addiction counsellors, social workers, nurses or psychologists, learned about personality disorders and the self-report instrument used in this study to assess personality disorder features, the DSM and ICD-10 Personality Questionnaire (DIP-Q). All patients participating were referred by local authorities for substance abuse treatment, and deemed in need of drug abuse treatment. Only patients from treatment units that served only drug abusers were included.
Staff members were instructed to hand out the DIP-Q to patients in their care. They were also requested to inform patients, that the data from the instrument would be used to both research purposes and in their own treatment, and that they would receive a personalized feedback on their test results. On the front of the DIP-Q, we added information about the potential use of the data from the questionnaire for research purposes. The professionals were not asked to include all patients in their units, thus the sample is a convenience sample. There are no institutional review boards in Denmark for research on human subjects that only includes psychosocial assessment or intervention. The research was carried out with respect for the Helsinki declaration.
The DSM-IV and ICD-10 Personality Questionnaire
The DSM-IV and ICD-10 Personality Questionnaire [DIP-Q] was used as the measure of personality pathology. The DIP-Q is a self-report questionnaire for screening for DSM-IV and ICD-10 PDs, plus schizotypal disorder in ICD-10 . The instrument is highly similar to other questionnaires measuring PDs, such as the SCID-IIQ, and the PDQ-R. It consists of 151 statements that must be rated as true or false, and three self-rating scales: severity of current events, global assessment of functioning, axis V of the DSM-IV for past year and global assessment of functioning for recent weeks. The DIP-Q was constructed through a consensus process. First, four psychiatrists selected a range of statements considered representative of the diagnostic criteria for each personality disorder. These statements could answered in a true/false format. The representative statements were then reviewed and validated by a second set of independent psychiatrists . A translation and an English version was made available from the Swedish authors. No details of this translation were available and therefore a new Danish translation was made based on the English and Swedish versions, and compared with the original translation.
Studies show indications of concurrent  and predictive  validity of the instrument. From the DIP-Q, we calculated the number of criteria in each of the three clusters of the DSM-IV, and for each personality disorder.
The Feeling Word Checklist-58
We used the Feeling Word Checklist-58 to measure emotional reactions to patients. The Feeling Word Checklist-58 (FWC-58) is based on the FWC developed by Whyte et al.  but expanded with 28 items – 23 items were feeling words that experienced therapists found were lacking in the original FWC, and five were taken from the PANAS scale developed by Watson & Lee . The new items were mainly connected to feelings of security, being invaded, idealized and devalued. It was developed in Norwegian, but translated from the English version and back-translated several times by the authors and several English native-speakers.
The instruction to the form is: When I am in conversations with patient ___ I feel ...". Each feeling word is rated on a 5-point likert scale ranging from "Not at all" to "very much". Røssberg and colleagues conducted a factor analysis of the instrument, and derived to superordinate factors and 7 lower-order factors . The two superordinate factors were labelled helpfulness and distance, and the lower-order factors were labelled important, confident, rejected, on guard, bored, overwhelmed and inadequate.
Power analysis showed that to detect correlations of 0.30 with an alpha of 0.05 and two tails, we needed 82 subjects to obtain a power of 0.80. We decided on 0.30 as a realistic correlation, based on previous studies of countertransferrence . We also conducted power analysis for a regression analysis with 3 predictors. We assumed that only one factor would be independently associated with each of the two dependent variables (see below). If the two covariates explained 1% of the variance, and the third covariate explained 10% of the incremental variance, the number needed to obtain a power of 0.80 was 74 with an alpha of 0.05.
We first conducted two regression analyses entering criteria count for each of the three DSM-IV clusters (A: Odd-eccentric, paranoid, schizoid, schizotypal; B: Dramatic-erratic, antisocial, borderline, histrionic, narcissistic; C: Anxious-fearful, avoidant, dependent, obsessive-compulsive), as predictors and the two main factors of the FWC as dependent variables in each analysis (i.e., helpfulness and distance). Symptom counts of all three clusters were entered simultaneously in the regression analyses. Before conducting these analyses we rank transformed all variables to reduce the impact of violations from normality, as several of the FWC scales had a small number of outliers.
In the next regression analyses, we regressed the higher order factors on all the Cluster diagnoses within any cluster with a significant impact on that factor, again entering all disorders in a cluster in one model. In a final step, we analyzed significant relationships by using the relevant FWC "small" (lower order) scales as dependent variables.
We also report the simple non-parametric Spearman correlations between all scales of the FWC, both the superordinate and the lower-order facets, and all DSM-IV criteria counts.