This paper describes an experiment to assess the anxiety responses of people giving 5 min. presentations to virtual audiences consisting of eight male avatars. There were three different types of audience behavior: an emotionally neutral audience that remained static throughout the talk, a positive audience that exhibited friendly and appreciative behavior towards the speaker, and a negative audience that exhibited hostile and bored expressions throughout the talk A second factor was immersion: half of the forty subjects experienced the virtual seminar room through a head-tracked, head-mounted display and the remainder on a desktop system. Responses were measured using the standard Personal Report of Confidence as a Public Speaker (PRCS), which was elicited prior to the experiment and after each talk Several other standard psychological measures such as SCL-90-R (for screening for psychological disorder), the SAD, and the FNE were also measured prior to the experiment. Other response variables included subjectively assessed somaticization and a subject self-rating scale on performance during the talk. The subjects gave the talk twice each to a different audience, but in the analysis only the results of the first talk are presented, thus making this a between-groups design. The results show that post-talk PRCS is significantly and positively correlated to PRCS measured prior to the experiment in the case only of the positive and static audiences. For the negative audience, prior PRCS was not a predictor of post-PRCS, which was higher than for the other two audiences and constant. The negative audience clearly provoked an anxiety response irrespective of the normal level of public speaking confidence of the subject. The somatic response also showed a higher level of anxiety for the negative audience than for the other two, but self-rating was generally higher only for the static audience, each of these results taking into account prior PRCS.
