Can a computer detect interpersonal skills? Using machine learning to scale up the Facilitative Interpersonal Skills task. (3rd April 2021)

• Record Type: Journal Article
• Authors: Goldberg, Simon B.; Tanana, Michael; Imel, Zac E.; Atkins, David C.; Hill, Clara E.; Anderson, Timothy
• Is Part Of: Psychotherapy research. Volume 31:Number 3(2021)
• Page Start: 281
• Page End: 288
• Publication Date: 2021-04-03
• Subjects: machine learning -- interpersonal skills -- Facilitative Interpersonal Skills -- therapist effects -- artificial intelligence; Psychotherapy -- Periodicals; Psychotherapy -- Research -- Periodicals; Psychotherapy -- Periodicals; Psychothérapie -- Périodiques; Psychothérapie -- Recherche -- Périodiques; 616.891405
• DOI: 10.1080/10503307.2020.1741047
• Languages: English
• ISSNs: 1050-3307

Abstract

ABSTRACT: Objective: Therapist interpersonal skills are foundational to psychotherapy. However, assessment is labor intensive and infrequent. This study evaluated if machine learning (ML) tools can automatically assess therapist interpersonal skills. Method: Data were drawn from a previous study in which 164 undergraduate students (i.e., not clinical trainees) completed the Facilitative Interpersonal Skills (FIS) task. This task involves responding to video vignettes depicting interpersonally challenging moments in psychotherapy. Trained raters scored the responses. We used an elastic net model on top of a term frequency-inverse document frequency representation to predict FIS scores. Results: Models predicted FIS total and item-level scores above chance ( rho s = .27–.53, p s < .001), achieving 31–60% of human reliability. Models explained 13–24% of the variance in FIS total and item-level scores on a held out set of data ( R 2 ), with the exception of the two items most reliant on vocal cues (verbal fluency, emotional expression), for which models explained ≤1% of variance. Conclusion: ML may be a promising approach for automating assessment of constructs like interpersonal skill previously coded by humans. ML may perform best when the standardized stimuli limit the "space" of potential responses (vs. naturalistic psychotherapy) and when models have access to the same data available to raters (i.e., transcripts).