Measuring Alliance Toward Embodied Virtual Therapists in the Era of Automated Treatments With the Virtual Therapist Alliance Scale (VTAS): Development and Psychometric Evaluation
Alexander Miloff, Per Carlbring, William Hamilton, Gerhard Andersson, Lena Reuterskiöld, Philip Lindner
Abstract
BACKGROUND: Automated virtual reality exposure therapies (VRETs) are self-help treatments conducted by oneself and supported by a virtual therapist embodied visually and/or with audio feedback. This simulates many of the nonspecific relational elements and common factors present in face-to-face therapy and may be a means of improving adherence to and efficacy of self-guided treatments. However, little is known about alliance toward the virtual therapist, despite alliance being an important predictor of treatment outcome. OBJECTIVE: In this study, we aimed to evaluate the first alliance instrument developed for use with embodied virtual therapists in an automated treatment format-the Virtual Therapist Alliance Scale (VTAS)-by (1) assessing its psychometric properties, (2) verifying the dimensionality of the scale, and (3) determining the predictive ability of the scale with treatment outcome. METHODS: A psychometric evaluation and exploratory factor analysis of the VTAS was conducted using data from two samples of spider-fearful patients treated with VRET and the help of an embodied, voice-based virtual therapist (n=70). Multiple regression models and bivariate correlations were used to assess the VTAS relationship with treatment outcome, according to self-reported fear and convergence with presence and user-friendliness process measures. RESULTS: The VTAS showed a sound two-factor solution composed of a primary factor covering task, goal, and copresence; adequate internal consistency; and good convergent validity, including moderate correlation (r=.310, P=.01) with outcomes over follow-up. CONCLUSIONS: These preliminary results suggest that alliance toward a virtual therapist is a significant predictor of treatment outcome, favors the importance of a task-goal over bond-factor, and should be explored in studies with larger sample sizes and in additional forms of embodiment.