Participants
The sample consisted of 93 participants (45 males and 48 females) with a mean age of 25.6 years (SD = 7.7, Range = 18–53). Males (M = 26.5 years) and females (M = 24.7 years) did not differ significantly with respect to age, t (86) = 1.12, p = .26. Scores on the Social Phobia Scale were lower for males (M = 17.1, SD = 9.68) compared to females (M = 22.7, SD = 12.7), and this difference reached statistical significance, t (91) = 2.36, p = .02.
The mean SPS score of the current sample was 20.0 (SD = 11.6, range = 2–48). Compared to McNeil et al.’s (1995) reference data, this is significantly lower than the mean SPS score of individuals with SAD, M = 32.8, SD = 14.8, t (57) = 5.86, p < .001, but significantly higher than undergraduates, M = 13.4, SD = 9.6, t (144) = 3.69, p < .001, and community volunteers, M = 12.5, SD = 11.5, t (141) = 3.70, p < .001. The mean age of these comparison groups was higher (SAD sample M = 36.5 years, community sample M = 33.2 years, with age data not reported for undergraduates) than the current sample.
An exclusion criterion of previous acquaintance with the experimenters was implemented, as familiarity may have reduced the effectiveness of the social challenge tasks as anxiety inductions. A recruitment request was e-mailed to all students at Greenwich University which stated that “volunteers are sought to take part in a paid (£10) study which will involve filling in some questionnaires, engaging in a conversation task and talking to others about a set topic, giving your views”.
Anxiety and social behavior scales
Mattick and Clarke’s Social Phobia Scale (SPS)Footnote 1 was used to assess level of trait social anxiety. The SPS consists of 20 items rated on a five-point (0–4) scale, with higher scores indicating greater social anxiety. The scale has been shown to reliably assess social anxiety in both non-clinical and clinical populations [23]. The SPS has previously demonstrated good test-retest reliability, internal consistency and convergent validity [24, 25] and exhibited high internal consistency (Cronbach’s α = .89) for the current data.
State anxiety was assessed in order to verify that the speech and interaction tasks resulted in increased anxiety relative to participants’ baseline anxiety. Baseline anxiety was assessed with a single self-report item that asked respondents to indicate their current anxiety on a scale of 1–10. State anxiety was also assessed immediately prior to the commencement of each task (participants had been provided with task details a few minutes earlier), and immediately after each task where participants were asked to rate the anxiety they had felt during the task itself. Single-item assessments of state anxiety have shown good reliability and convergent validity [26].
The Social Performance Rating Scale (SPRS) [27] was used to rate the participant on the following five dimensions: Gaze - adequacy of eye contact, Vocal Quality – warmth, clarity and enthusiasm demonstrated in verbal expression, Length – low level of monosyllabic speech/excessive talking, Discomfort – low levels of behavioral anxiety (e.g., fidgeting, trembling, postural tension), and Flow - verbal fluency (including the ability to incorporate information provided by the conversation partner smoothly into the interaction). The flow item was not used in the assessment of the speech task, as the rating descriptors for this component are specific to conversation. All SPRS items were rated on a 5-point scale and scored so that higher scores represented more effective social performance. Detailed descriptive anchors accompany each rating point to facilitate scoring; for example, Vocal Quality, “5 (Very Good) = Participant is warm and enthusiastic in verbal expression without sounding condescending or gushy”. The SPRS has shown excellent inter-rater reliability, internal consistency, convergent, discriminant and criterion validity [27, 28]. Agreement across the three raters assessing the speech task was examined with an intraclass correlation (ICC). An absolute-agreement model was used [29], which is a stringent test requiring both high inter-rater correlations and minimal discrepancy in actual rating values to produce a high ICC. Analysis revealed ICC’s = .64–.86 for individual SPRS dimensions (all p’s < .001), suggesting good rater agreement [30]. Scores were therefore averaged across raters for each individual SPRS dimension for the speech task. Similar means (range: 3.4–3.8) and standard deviations (range: 0.7–1.1) were observed across SPRS components for both interaction and speech tasks.
Speech task
Participants were given 3 min to prepare a speech presenting a persuasive argument on their choice of one of the following topics: “sometimes it is ok to lie, discuss” or “can any crime be justified?”. Participants were told they would be presenting in front of a small audience and that they should try to keep going for 3 min although they could terminate the task at any point. Three confederates (one male and two female) comprised the “audience” for the speech task, with the same three-confederate audience used for each participant. The confederate audience had previously undertaken a number of trial sessions with several undergraduate volunteers acting as participants where they had practiced maintaining neutral facial expressions.
Interaction task
Participants were told that they would shortly be introduced to someone and that they would have 3 min to find out as much as they could about this person, although they could terminate the task at any time. The conversation partner was an experimental confederate, who was of the opposite-sex in order to maximize socially-evaluative challenge [6]. The same male confederate was used for each female participant, and the same female confederate was used for each male participant, with the one male and one female confederate taken from the pool of three confederates used in the speech task. Confederates had previously undertaken a number of trial sessions amongst each other and with undergraduate volunteers, where they practiced giving minimal responses, avoiding asking questions and maintaining neutral facial expressions [6]. Nobody other than the participant and the confederate was present during the interaction task when the experiment began.
Procedure
To put participants in a relaxed state for a reliable assessment of baseline state anxiety, and to provide time for the experimenter to prepare the social challenge tasks, participants watched a 5-min relaxation video showing images of various seascapes accompanied by relaxing sounds. They then immediately completed the baseline state anxiety item along with the Social Phobia Scale and were randomized to undergo either the speech or interaction task first.
Participants were given details of the first social challenge task and reminded that they had the right to withdraw from the study at any point (no withdrawals occurred). Immediately prior to the social challenge task, participants completed the state anxiety item to assess anticipatory anxiety. Immediately following the task, participants again completed the state anxiety item, retrospectively indicating the anxiety they had experienced during the task. Participants were independently rated on their social performance by the audience of confederates (speech task) or the conversation partner (interaction task) using the SPRS, with ratings not disclosed to participants. This procedure was then repeated with the second social challenge task.
Statistical analysis plan
The association of social anxiety and sex with observer ratings was examined by conducting separate regression analyses on each SPRS dimension, with predictors of social anxiety, sex (− 1 = males, + 1 = females) and a Social Anxiety X Sex interaction term. Social anxiety was standardized but SPRS ratings were left unstandardized, so that the raw regression coefficient is interpreted as the mean change in rating points (on the 1–5 scale) following a one standard deviation increase in social anxiety. The interaction term was computed by cross-multiplication of sex and standardized social anxiety scores [31].
To determine whether regression coefficients of social anxiety and behavioral ratings differed significantly across the different SPRS dimensions, we tested the equality of these coefficients within a structural equation model. Predictors were the same as for the multiple regression analysis described above, and outcome variables were two SPRS dimensions (specified with correlated errors) whose coefficients were to be compared. We then imposed an equality constraint on the coefficient of social anxiety with each of two performance dimension coefficients. If a likelihood ratio test indicates a significant decrease in fit when an equality constraint is used, this indicates that the two coefficients are not equal [32]. Analyses were conducted in R using the lavaan [33] package .