H567/02 Section A - Hancock et al. (2011): language of psychopaths

The question and predicted language patterns

The researchers asked whether psychopathy-related characteristics would be reflected in the language of homicide offenders describing their crimes.

The study uses the term psychopathy for a pattern assessed with the Psychopathy Checklist-Revised, including interpersonal/affective features and an impulsive or antisocial lifestyle. It is not synonymous with psychosis, autism or all mental ill health. The prescribed study is Hancock et al. (2011), its first-publication identity; the final journal issue appeared in 2013.

The researchers predicted three broad patterns. An instrumental, goal-directed view of the crime might appear in more explanatory cause-and-effect words such as “because”. A focus on basic material needs might appear in more references to food, drink and money and fewer to family or religion. An emotional deficit might appear in less affective language, more speech disfluencies such as “um”, and linguistic distancing through past-tense rather than present-tense verbs.

Word choice was treated as an indicator of underlying cognition and emotion, not a transparent record of thought. Many things can affect a narrative, including the event, memory, the interview and attempts to present oneself favourably. The study belongs to individual differences because it compares psychological profiles, and to measuring differences because it operationalises language features and relates them to an assessed characteristic.

Participants and psychopathy assessment

The sample comprised 52 male homicide offenders in Canadian correctional facilities who admitted their crime and volunteered: 14 classified as psychopathic and 38 as non-psychopathic for this research.

This was a comparison of pre-existing characteristics, using interviews and content analysis. The researchers did not manipulate psychopathy or randomly allocate people to it. The groups did not differ significantly in age or elapsed time since the homicide. Those checks reduce particular alternative explanations, but “not significant” does not prove groups identical on every relevant characteristic.

The PCL-R has 20 criteria, each scored from zero to two, producing a maximum of 40. Factor 1 concerns interpersonal/affective features, while Factor 2 concerns an impulsive and antisocial lifestyle. The paper describes a clinical cut-off of 30 but used a previously justified research cut-off of 25 to classify this sample. Do not replace the study's rule with 30 or treat its research classification as a universal clinical rule.

Assessments came from trained prison psychologists for 39 files and a trained researcher for 13. A trained graduate student independently re-coded ten randomly selected files. The reported positive correlation was r = .94, p < .001, supporting inter-rater consistency in scores. It does not show perfect agreement or prove that language itself is a diagnostic measure.

Volunteers who admitted their crimes may differ from offenders who deny responsibility or decline interviews. All participants were men in Canadian prisons; the comparison group was other homicide offenders, not the general public. Findings therefore cannot automatically describe women, non-offending people with psychopathic traits or everyone in another culture.

From interviews to linguistic measures

Participants gave detailed crime accounts in recorded, open-ended interviews using a standardised Step-Wise prompting procedure, lasting about 25 minutes.

The purpose and procedure were explained verbally. Interviewers—two senior psychology graduate students and one research assistant—were blind to the participants' psychopathy scores. This reduced the opportunity for score-based expectations to influence prompting. Participants were encouraged to describe the offence from beginning to end without leaving details out.

Transcripts were close to verbatim, preserving disfluencies such as “uh” and “um”. Spelling was checked, interviewer speech removed, and proper nouns and abbreviations expanded. Preserving the disfluencies was essential because fluency was a dependent measure; deleting them would systematically alter that outcome.

Wmatrix classified parts of speech and semantic categories, such as grouping money-related words together. It compared the combined corpus of speech from the 14 classified participants with the combined control corpus. Dictionary of Affect in Language (DAL) assessed pleasantness, intensity and imagery for each participant's narrative. Corpus-level category comparisons and individual-level affect scores are different units of analysis.

The original transcripts are qualitative accounts. Coding turns selected features into quantitative frequencies or proportions, enabling comparisons but losing some of the individual story and context. A raw count must be interpreted relative to the amount of language: 100 references in 10,000 words is a different rate from 100 in 100,000. The percentage of corpus words is category frequency divided by total corpus words, multiplied by 100: