Master's thesis (n=20): no significant physiological or subjective anxiety differences between adult males who stutter and matched non-stuttering controls during VR speech

Background

This 2011 master’s thesis by James McKay Brinton, supervised by Shelley Brundage at The George Washington University, set out to integrate physiological and subjective measures of anxiety in adult males who stutter (AWS) and matched non-stuttering males (AWNS) during ecologically valid VR speaking tasks. Prior research on the relationship between stuttering and anxiety had been “distinct but inconsistent” (Bloodstein, 1995): some studies found AWS to show greater anxiety than AWNS on physiological or self-report measures during speaking tasks, others found no group differences. The conflicting literature had been attributed by Weber & Smith (1990) and others to inconsistent task selection, participant selection from clinic waiting lists (which may over-represent high-anxiety individuals), and over-reliance on a single anxiety measure.

Brinton’s thesis aimed to address those methodological gaps by combining three physiological measures (galvanic skin response GSR, heart rate HR, respiration rate RESP) with a self-report measure (Subjective Units of Distress Scale, SUDS) in an ecologically valid VR public-speaking context, and to compare AWS with strictly age- and gender-matched non-stuttering controls.

What the researchers did

Ten adult males who stutter (AWS, mean age 30.8 SD=12.4, ages 18-51) and ten age-matched (within 3 years) non-stuttering males (AWNS) were recruited from the Metropolitan Washington, DC area. AWS participants came from speech and hearing clinics and local National Stuttering Association support groups; AWNS were recruited by word of mouth, campus flyers, and snowball sampling. All AWS were professionally verified as persons who stutter and ranged from very mild to severe (SSI-4 10-35, mean 20.9 SD=10.6). Both groups were excluded for general anxiety disorders, speech/language disorders other than stuttering, motion sickness, epilepsy/seizure disorders, and diagnosed psychiatric disorders with accompanying medication.

Equipment. VR was delivered via an eMagin Z800 head-mounted display + noise-cancelling headphones, driven by a Dell Precision 390 desktop. Three virtual reality environments (VREs) developed by Virtually Better, Inc. (Decatur, GA) were used: (1) an orientation environment with the participant seated in a virtual parked car with ambient bird/traffic sounds (for initial VR orientation and a 60-second silence baseline), (2) a virtual audience of approximately 30 listeners (mixed gender and ethnicity) seated in five rows in a medium-sized classroom-like room with the participant standing behind a virtual desk, and (3) an empty version of the same room with empty chairs. The researcher could trigger audience reactions (whispering, yawning, puzzled facial expressions, falling asleep, slight head-nodding) via keyboard hot keys.

Physiological data were collected with a BIOPAC MP150 system + AcqKnowledge 4.1 software via a Dell Inspiron 1525 laptop. ECG/HR electrodes on right wrist and left ankle; GSR electrodes on the middle and index fingers of the right hand; an elastic respiration belt around the chest. A Valsalva maneuver was performed prior to tasks to confirm GSR equipment was working. Audio was captured with an Olympus WS-500M digital recorder with lapel microphone.

Tasks. All participants completed 10 counter-balanced tasks (including high-stress and low-stress speaking, and silence). The thesis’s primary research questions covered three tasks: (a) 60 seconds of silence in the VR car (orientation); (b) two 4-minute speeches, one to the virtual audience and one to the empty-chairs version of the same room (counter-balanced order); and (c) reading the Grandfather Passage aloud (low-stress condition, with 30-second silence periods between repetitions that were combined into a 60-second non-VR silence baseline for comparison with the VR-silence period). Speech topics were everyday topics (favorite restaurants, sports, hobbies, books). SUDS ratings were collected after each minute of each speech. A Presence Questionnaire-Revised (Witmer & Singer, 1998) was completed after the protocol.

Self-report battery (pre-task). Trait STAI (STAI-T), Fear of Negative Evaluation (FNE), Modified Erickson Scale of Communication Attitudes (S-24), and a 5-point speaking-related anxiety scale. AWS also completed the OASES.

Analysis. Multivariate analyzes of variance (MANOVA) in SPSS 15.0 with setting (audience vs empty chairs) and group (AWS vs AWNS) as independent variables and GSR, HR, RESP, and SUDS ratings as dependent variables. Significance level α=.05. Inter-rater reliability of biophysical data entry was 98.9% for GSR, 97.9% for HR, and 94.3% for RESP. SSI-4 inter-rater reliability was 90%.

What they found

Between-group differences (the primary research questions). A repeated-measures MANOVA found NO significant between-group differences between AWS and AWNS on any physiological measure during silence tasks (GSR F(1,18)=1.69 p=.21; HR F(1,18)=4.51 p=.05; RESP F(1,18)=1.83 p=.19) or during the two VR speaking tasks (overall F(3,16)=2.924 p=.066; individual measures all p>.19). No significant between-group difference in SUDS ratings during the VR speaking tasks. No significant interaction effects between group and setting (F(3,16)=.318 p=.81 for silence; F(3,16)=1.755 p=.196 for speaking).

Within-group effects. A significant within-group effect of setting on SUDS emerged: both groups reported significantly higher SUDS in the audience condition than the empty-chairs condition (F(1,18)=11.2, p=.004). Physiological measures (GSR, HR, RESP), in contrast, did NOT differ significantly between audience and empty-chairs settings within either group. Heart rate during silence WAS significantly higher in the read-silence period than the VR-silence period (F(1,18)=13.077, p=.002) within groups, interpreted as elevated arousal carried over from surrounding speaking tasks rather than a property of the VR environment per se.

Time course. SUDS did not change significantly across the four one-minute time points within a 4-minute speech (F(3,51)=2.60, p=.062), so post-1-minute SUDS values were used as the dependent variable in the between-group analyzes.

Baseline trait/speaking-anxiety measures (pre-task). Although the within-task measures did not differentiate groups, pre-task self-reports DID: AWS scored higher than AWNS on the 5-point speaking-related anxiety scale (M=2.6 SD=.74 vs M=1.6 SD=.84), on the Erickson S-24 (M=14.1 SD=4.84 vs M=5.8 SD=3.55), and slightly higher on FNE (M=12.8 SD=8.05 vs M=11.2 SD=10.91) and STAI-T (M=54.9 SD=3.57 vs M=53.3 SD=3.53). The thesis does not statistically compare these baseline measures between groups but reports them descriptively to characterize the sample.

Why this matters

For the Evidence Hub, this thesis is the methodological precursor to Brundage, Brinton & Hancock (2016) in Journal of Fluency Disorders (already in the Evidence Hub), which expanded and refined the analyzes presented here. Both works converge on the same central observation: in a virtual audience condition, subjective distress is reliably elevated above the empty-room baseline while physiological measures do not show a corresponding setting-specific elevation. The dissociation between subjective and physiological channels means clinicians using VR exposure should collect both: relying on autonomic biofeedback alone can lead to the conclusion that an exposure was uneventful when in fact the client experienced substantially elevated distress.

The null between-group finding (AWS vs AWNS) is also clinically informative. It cautions against assumptions that adults who stutter routinely show greater within-task anxiety responses than non-stuttering peers in VR speaking contexts. The fact that AWS scored higher than AWNS on pre-task trait/speaking-anxiety self-reports BUT did not differ on within-task SUDS or physiological responses suggests that trait/dispositional measures and state/within-task measures may capture different aspects of the AWS experience.

Limitations

The thesis author flags the following limitations explicitly:

• Low statistical power. Page 30 acknowledges “low statistical power possibly due to small sample size, low effect size, and large variance in measures of anxiety.” With n=10 per group, the study was underpowered to detect small or medium effects.
• All-male sample. Both groups consisted exclusively of males (10 per group); the thesis does not comment on this generalizability constraint, but it is a clear limit on transferability to female adults who stutter.
• Master’s thesis, not peer-reviewed journal publication. The thesis was completed in May 2011 as a Master of Arts degree requirement and was not directly published in a peer-reviewed venue. A related/expanded analysis was later published as Brundage, Brinton & Hancock (2016) in Journal of Fluency Disorders.
• Single session, no follow-up. The thesis assessed responses to a single VR exposure session; it does not address whether responses change with repeated exposures or over time.
• Recruitment from clinic waiting lists and support groups. As the thesis itself acknowledges (citing Craig et al. 2003; Iverach et al. 2009), recruitment from clinic waiting lists and support groups may produce samples with higher anxiety than the general PWS population, potentially limiting generalizability.
• Brundage lab / Virtually Better, Inc. background. Not flagged as a limitation by the thesis itself, but readers of the Evidence Hub should be aware that this thesis was completed in the Brundage lab at GWU using VRE software developed by Virtually Better, Inc., the same commercial VR-software vendor that featured in Brundage’s earlier NIH SBIR-funded work (Brundage et al., 2006; Brundage, 2007). This is relevant background context for evaluating the line of work.
• Speech topics not standardized across participants. Topics were everyday topics (favorite restaurants, sports, hobbies, books) but were given as needed rather than fully standardized, which may have contributed to between-participant variability.