VR job interviews show interviewer style affects stuttering frequency; %SS in VR correlates strongly with %SS in a clinical SSI-3 interview

Background

Generalization of treatment gains from the clinic to real-world settings is a persistent challenge in stuttering therapy. Taking clients into the situations they actually fear (job interviews, presentations, group meetings) is often impractical and risks compromising client confidentiality. Role-play with a familiar clinician rarely captures the genuine stress of speaking with a stranger in a high-stakes situation. The authors propose that virtual reality environments (VREs) could fill this gap by providing controllable, repeatable, confidential interim steps between the therapy room and the real world. This study set out to test whether one such environment - a virtual reality job interview - could elicit measurable changes in stuttering frequency as a function of interviewer communication style.

What the researchers did

Twenty-three adults who stutter were recruited from speech and hearing clinics and stuttering support groups in the Washington, DC area. Three were excluded (one recorder failure, one severe-stuttering participant could not complete the tasks in the 45-minute time limit, one late arrival), leaving 20 participants (6 women) whose data were analyzed. Stuttering severity using the Stuttering Severity Instrument 3rd edition (SSI-3) ranged from 10 to 43 (mean 21.65), spanning very mild to severe.

The Virtual Reality Job Interview (VRJI) was developed by the authors in collaboration with Virtually Better, Inc. and consisted of an elevator, waiting room, hallway, and two interview offices. The challenging interview took place in the larger CEO office (decorated with leather chairs, wooden furniture, and diplomas); the supportive interview took place in a smaller human-resources office (decorated with metal furniture). Interviewer behavior was varied between conditions: the challenging interviewer interrupted, spoke faster (~320 syllables per minute), broke eye contact, and reacted to answers with a sarcastic tone and facial expressions of confusion; the supportive interviewer maintained eye contact, did not interrupt, opened with the disclosure “I might stutter during this interview…”, and spoke at ~270 syllables per minute. Interview order and interviewer gender were counterbalanced.

Equipment: Dell P-IV PC, VFX-3D head-mounted display (640×480 resolution per eye), head tracker, and a ScentPalette olfactory simulator (coffee in the waiting room, leather in the CEO office, cleaning supplies in the hallway). Each VR interview lasted approximately 20 minutes; total time in VR did not exceed 45 minutes per participant. Investigator-controlled “hot keys” allowed the virtual interviewer’s reactions to be varied within a fixed set of pre-recorded responses (the people in the VRE were video-integrated images of live actors).

Outcomes measured: percentage of stuttered syllables (%SS, as defined by Yaruss, 1997) during participants’ responses to two “board member” questions in each VR interview; %SS during the SSI-3 clinical interview task completed prior to VR exposure; Personal Report of Communication Apprehension-24 (PRCA-24); Personal Report of Confidence as a Speaker (PRCS); and a 19-item adapted Presence Questionnaire (Witmer & Singer, 1998) completed post-VR. Inter-judge agreement on disfluency coding was 81% after resolution.

What they found

%SS was significantly higher in the challenging VR interview than the supportive one (mean 8.71 vs 7.55; one-tailed t = 1.72, p = 0.05 across all 20 participants; t = 2.14, p = 0.02 when omitting the four participants with severe stuttering whose %SS was 4-8 times higher than the rest of the sample). Interviewer communication style therefore appeared to influence stuttering frequency inside the virtual environment.

%SS in both VR conditions correlated very strongly with %SS during the SSI-3 clinical interview task: r = 0.904 (p = 0.0001) for challenging VR vs SSI-3, and r = 0.930 (p = 0.0001) for supportive VR vs SSI-3. The authors interpret this as evidence that “the frequency of stuttering behaved in similar ways during interviews with virtual and real people.” The comparison was between VR and a clinical interview task, NOT between VR and a real-world job interview - the paper is explicit that direct VR-vs-real-world comparisons were ongoing future work at the time.

Stuttering severity was not significantly correlated with sense of presence (r = 0.273, p = 0.25), speaking confidence (PRCS, r = 0.364, p = 0.11), or speaking apprehension (PRCA-24, r = -0.127, p = 0.59). Qualitative debrief comments suggested participants experienced realistic emotional and physical reactions (“I felt very stressed out”; “I felt fear and avoidance about stuttering”; “When I was walking in, my hands were nervous”); some reached out their hands to shake the virtual interviewer’s hand despite being seated throughout.

Why this matters

This was one of the earliest empirical demonstrations that a controllable virtual job interview could elicit measurable, condition-dependent changes in stuttering frequency in adults who stutter, and that those frequencies aligned closely with stuttering measured during a separate clinical interview task. The paper helped open the door for VR to be used as an interim step between the therapy room and the real world. It did NOT establish that VR practice transfers to real-world job interviews, and the authors do not claim that it does; they identify this as the next research question.

Limitations

The authors explicitly flag the following in their discussion:

• VR-vs-real-world comparison was not conducted. The strongest correlations were with a separate clinical interview task (the SSI-3), not with a real-world job interview. The authors note that direct VR-vs-real-world comparisons were “underway” but not reported in this paper.
• Severe stuttering under-represented. Only 4 of the 20 analyzed participants had severe stuttering; the authors flag this as a future-research priority (“validate the use of VR with a larger sample of participants with severe stuttering”).
• One speaking situation tested. Only a job-interview scenario was used; other workplace-related interactions (small group presentations, brainstorming sessions) were noted as planned future development.
• Virtual interviewers were investigator-controlled within a fixed set of pre-recorded responses. The interviewer’s reactions could be varied via keyboard hot keys, but the conversation was not generatively responsive to what the participant said. This limits ecological realism of the interaction itself.
• No physiological data collected. The authors note plans to add heart rate, galvanic skin response, and cortisol in future studies (the cortisol work followed in Duncko, Brundage, Graap, Kling, & Gold, 2006).
• Initial hardware cost and motion-sickness risk were flagged by the authors as practical barriers; total VR exposure was capped at 45 minutes per participant to mitigate simulator sickness.
• No assessment of long-term outcomes. The study measured a single VR exposure session per condition; it does not address whether repeated VR practice changes communication outcomes over time.