Introduction
Social anxiety (SA) is characterized by ‘a marked fear or anxiety about one or more social situations in which the individual is exposed to possible scrutiny by others’ (DSM-V-TR; American Psychiatric Association, 2022). SA is a psychological problem that can cause a marked deterioration in social functioning (Halls et al., Reference Halls, Cooper and Creswell2015). The prevalence rates of SA are high, especially in young people, who often encounter social stressors in both academic and social contexts (de Lijster et al., Reference de Lijster, Dieleman, Utens, Dierckx, Wierenga, Verhulst and Legerstee2018). Specifically, between 27 and 36% of adolescents and young people reported high levels of SA (Jefferies and Ungar, Reference Jefferies and Ungar2020; Mekuria et al., Reference Mekuria, Mulat, Derajew, Mekonen, Fekadu, Belete, Yimer, Legas, Menberu, Getnet and Kibret2017).
Cognitions play an important role in the initiation and maintenance of SA (Ledley and Heimberg, Reference Ledley and Heimberg2006; Spence and Rapee, Reference Spence and Rapee2016). In fact, the theoretical model of SA holds that cognitive biases are frequent in people with SA (Clark and Wells, Reference Clark and Wells1995; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014). When people with SA are exposed to a social event, various types of cognitive vulnerabilities, such as interpretation bias, could lead to a more negative interpretation of social stressors and thus increase the probability of reacting with high levels of SA.
Interpretation bias as cognitive bulnerability to social anxiety
Interpretation bias is one of the cognitive vulnerabilities that has been the focus in several SA models (Clark and Wells, Reference Clark and Wells1995; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014). It has been described as ‘a tendency to interpret ambiguous information in a consistent manner, which is usually threatening or negative’ (Schoth and Liossi, Reference Schoth and Liossi2017, p. 1). Thus, in an ambiguous situation, such as being observed by someone at a party, a person with interpretation bias would tend to interpret the situation negatively. Possibly, the individual could think that people around them are evaluating or speaking in a harmful manner about him or her.
Interpretation bias contributes to experiencing ambiguous situations as negative, mildly negative, or catastrophic (Stopa and Clark, Reference Stopa and Clark2000). They have mainly been evaluated using two well-differentiated approaches. The first method is based on the evaluation of the interpretations that the participants make about ambiguous faces (Gutiérrez-García and Calvo, Reference Gutiérrez-García and Calvo2017; Gutiérrez-García et al., Reference Gutiérrez-García, Fernández-Martín, Del Líbano and Calvo2019; Maoz et al., Reference Maoz, Eldar, Stoddard, Pine, Leibenluft and Bar-Haim2016).
Generally, several images of faces with different ambiguity levels and different emotions are created. These images are all presented to participants and rated by them. In general, studies using this method have indicated a greater tendency towards interpretation bias with ambiguous faces in people with SA. The second method is based on the imagination of ambiguous social scenarios, where participants are asked about what kinds of interpretations they would make in those situations (Miers et al., Reference Miers, Blöte, Bögels and Westenberg2008; Miers et al., Reference Miers, Sumter, Clark and Leigh2020). These types of studies have shown that negative interpretations are more frequent in people with high levels of SA. In fact, a meta-analysis that included a total of 44 studies with both methodologies found a large effect size (g = 0.83) in the relationship between interpretation bias and SA (Chen et al., Reference Chen, Short and Kemps2020).
According to several theoretical models (Clark and Wells, Reference Clark and Wells1995; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014), it is expected that individuals with a greater tendency towards interpretation bias will experience higher levels of SA when faced with a social stressor. In fact, interventions based on reducing interpretation bias reduce SA levels in response to social stressors (Hoppitt et al., Reference Hoppitt, Illingworth, MacLeod, Hampshire, Dunn and Mackintosh2014). However, although both social stressors (Auerbach et al., Reference Auerbach, Richardt, Kertz and Eberhart2012) and interpretation bias in ambiguous situations (Chen et al., Reference Chen, Milne, Dayman and Kemps2019) predict higher levels of SA, no studies have examined whether interpretation bias moderates the association between social stressors and SA symptoms.
The role of safety behaviours in social anxiety
Safety behaviours (Safe-B) are a relevant maladaptive strategy in the anxiety context (Salkovskis, Reference Salkovskis1991). According to several models of SA, when faced with a negative evaluation of a social stressor, people with SA who do not escape or avoid the situation tend to engage in Safe-B (Clark and Wells, Reference Clark and Wells1995; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014; Rapee and Heimberg, Reference Rapee and Heimberg1997). These have been described as the attempts made by people with anxiety to prevent or avoid the adverse outcomes of the threat (Piccirillo et al., Reference Piccirillo, Taylor Dryman and Heimberg2016). For example, with the aim of reducing their anxiety level or the likelihood of being evaluated by others, individuals with social anxiety tend to engage in certain behaviours, such as avoiding looking into the eyes of others or talking as little as possible.
Many studies have demonstrated the relationship between Safe-B and SA (e.g. Chiu et al., Reference Chiu, Clark and Leigh2021; Kocovski et al., Reference Kocovski, MacKenzie, Albiani, Battista, Noel, Fleming and Antony2016; Moscovitch et al., Reference Moscovitch, Rowa, Paulitzki, Ierullo, Chiang, Antony and McCabe2013). Individuals with SA employ Safe-B intending to momentarily reduce the anticipated negative consequences in the social scenario – mainly the perception of anxiety and negative evaluation (McManus et al., Reference McManus, Sacadura and Clark2008). However, performing Safe-B could perpetuate SA for several reasons. First, engaging in Safe-B would make it difficult to obtain evidence to disconfirm the effect caused by the social situation, nor would it allow the extinction of the anxiogenic response itself (McManus et al., Reference McManus, Sacadura and Clark2008; van Uijen et al., Reference van Uijen, Dalmaijer, van den Hout and Engelhard2018). In addition, the person with SA would attribute the prevention of the feared outcome to their own Safe-B (Piccirillo et al., Reference Piccirillo, Taylor Dryman and Heimberg2016).
However, although Safe-B and interpretation bias have been found to be related to SA, the nature of the relationship between Safe-B and interpretation bias with respect to SA in social scenarios remains to be elucidated. Both the tendency towards negative thoughts and Safe-B seem to independently predict SA levels (Chiu et al., Reference Chiu, Clark and Leigh2021). A cross-sectional study found that the association between interpretation bias in ambiguous scenarios and SA could be explained by Safe-B (Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Miers and Calvete2022a). Other cross-sectional studies have found that cognitive biases could explain the relationship between Safe-B and SA (Korte et al., Reference Korte, Unruh, Oglesby and Schmidt2015; Viana and Gratz, Reference Viana and Gratz2012).
However, the cross-sectional methodology applied in the latter studies did not allow establishing the direction of the relationships. A recent longitudinal study suggested an indirect relationship between interpretation bias measured with a computerized task and SA through Safe-B (Prieto-Fidalgo and Calvete, Reference Prieto-Fidalgo and Calvete2023). In that study, interpretation bias did not explain the indirect relationship between Safe-B and SA.
Although longitudinal studies that have examined whether Safe-B mediate the predictive association between interpretation bias and SA are scarce, this hypothesis has been evaluated for other negative cognitive styles. For example, in a longitudinal study involving evaluation through diaries, Safe-B were found to mediate the relationship between self-portrayal and SA (Moscovitch et al., Reference Moscovitch, Rowa, Paulitzki, Ierullo, Chiang, Antony and McCabe2013). Another longitudinal study showed that the lack of perceived anxiety control led to a greater number of Safe-B, which in turn led to higher levels of anticipatory anxiety before a talk (Carnahan et al., Reference Carnahan, Carter and Herr2020).
Current study
Current theoretical models of SA maintain that interpretation bias and Safe-B play a relevant role in the development and maintenance of SA (Leigh and Clark, Reference Leigh and Clark2018). However, there is a significant gap in understanding of the mechanisms involved, partly due to the scarcity of longitudinal studies. The first question addressed by this study was whether Safe-B acts as a mediating mechanism between interpretation bias and social stressors, on the one hand, and SA, on the other. The second question was whether interpretation bias moderates the association between stressors and SA. Finally, we examined whether prior levels of SA would predict higher Safe-B use, especially in co- occurrence with stressors.
To examine these questions, an intensive longitudinal design was utilized. Intensive longitudinal methods involve sufficient repeated measurements to permit the characterization of the change process for each subject (Bolger and Laurenceau, Reference Bolger and Laurenceau2013). This method makes it possible to focus on daily events and to make conclusions about within-subject and between-subject hypotheses. One of the fundamental benefits of this method is that it enables the assessment of thoughts and behaviours in a natural context (Bolger and Laurenceau, Reference Bolger and Laurenceau2013).
The proposed hypotheses are displayed in Fig. 1. Regarding the first question, previous studies indicate that people with high levels of SA perform Safe-B to a greater extent than those with low levels (Kocovski et al., Reference Kocovski, MacKenzie, Albiani, Battista, Noel, Fleming and Antony2016; Leigh et al., Reference Leigh, Chiu and Clark2021), Safe-B contribute to maintaining SA (Clark and Wells, Reference Clark and Wells1995; Leigh and Clark, Reference Leigh and Clark2018), and the association between interpretation bias and SA can be explained by Safe-B (Prieto- Fidalgo and Calvete, Reference Prieto-Fidalgo and Calvete2023). Therefore, we expected that daily social stressors (H1) and interpretation bias (H2 and H3) would be associated with daily Safe-B and that Safe-B, in turn, would be associated with daily SA (H4) and SA at the follow-up (H5).
The hypothetical multi-level structural equation models to be tested. The dotted lines represent moderation paths. Paths marked as ‘c’ are included as control paths.
Additionally, we expected that stressors would be associated directly with daily SA (H6), and interpretation bias would predict directly daily SA (H7 and H8) and SA at the follow-up (H9 and H10).
Regarding the second question, we examined the role of two modalities of interpretation bias (faces interpretations and scenarios interpretation). As predicted by theoretical models of SA, individuals with cognitive biases would be prone to more SA when experiencing social stressors (Clark and Wells, Reference Clark and Wells1995; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014). Thus, we hypothesized that face and scenario interpretations would increase the relationship between daily social stressors and daily Safe-B (H11 and H12) as well as between daily social stressors and daily SA (H13 and H14).
Finally, regarding the third question, given that levels of SA can play a determining role in daily experiences, we expected that initial SA would predict both daily Safe-B (H15) and SA at the daily level (H16) and at follow-up (H17). Moreover, we hypothesized that initial SA levels would increase the association between daily social stressors and Safe-B (H18) and daily SA (H19).
Method
Participants
The participants in this study were a subsample of a larger sample (n = 842) of vocational training students. They were invited to participate in a 7-day daily assessment, and 322 were interested in participating. Among these, balancing gender and SA level (see ‘Design and procedure’ section below), 150 students were selected. Twelve participants did not respond at least five times (days), and their data were discarded. Thus, 138 vocational training students participated, with a mean age of 20.15 years (SD = 2.5). About half of the students (49%) were women (n = 67). Of the 139 students who completed the diaries, 116 answered a follow-up measure. The larger sample was also used in two measures validation studies (Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Miers and Calvete2022a; Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Mueller and Calvete2022b) and a longitudinal study (Prieto-Fidalgo and Calvete, Reference Prieto-Fidalgo and Calvete2023). However, the 7–day daily assessment data were not employed in any of the above studies.
Design and procedure
An intensive longitudinal design (Bolger and Laurenceau, Reference Bolger and Laurenceau2013) was arranged in three steps: first, an initial measure was collected; second, a 7-day daily assessment was carried out; and, third, a 2-month follow-up assessment was conducted. Nine vocational centres in Bizkaia (Spain) agreed to collaborate in the study, giving access to the participants.
In order to collect the initial data, the researchers visited the education centres. After being informed about the study procedure, including the 7-day daily assessment, the participants answered the battery of scales using Qualtrics® on a computer. At the end of the survey, the participants expressed their interest in collaborating in the 7-day daily assessment. In this step, the participants were asked to provide their mobile phone number and email address so they could be sent the daily assessment. This information was saved in a different database, where a key was used to associate the phone number and email address with the raw data. In order to maintain privacy, this key was deleted after the selection process explained below.
One hundred and fifty participants were selected, considering the sex and SA levels of the initial measures. Specifically, balancing sex, participants with the lowest and highest levels of SA were selected. All the participants with the lowest levels of SA reported a SA score below 37 points and, except for seven participants, all participants with the highest levels of SA reported a SA score over 44 points. These are, respectively, the cut-offs recommended by the Spanish version of the SAS-A (Olivares et al., Reference Olivares, Ruiz, Hidalgo, García-López, Rosa and Piqueras2005) to identify ‘non-socially anxious’ (<37) individuals and to detect SA (>44). Thus, although the sample was not clinical, the sample used reported scores consistent with those obtained in clinical samples.
The survey was developed in a mobile-responsive way to improve the user experience when answering questions. A platform developed for this purpose sent an automatic message through WhatsApp at 6 p.m. daily. If the participant did not answer, a reminder was sent at 9 p.m. If a participant had not answered the seven diaries on the seventh day, one or two extra reminders were sent for the next two days. Therefore, all participants who answered at least five diaries completed them within the 9-day deadline. Specifically, 87% of the participants completed seven diaries on seven consecutive days. The rest of the participants needed one or two extra days. Except for three participants who completed diaries for the minimum number of days required for the data analysis (five diaries), the rest answered for seven days. Participants who completed at least five diaries were rewarded with a voucher worth 5 euros for an online sales company.
The data collection process guaranteed the privacy of the participants’ data. The platform, which managed the message sending, generated a personalized link with a private token. When the participants finished answering the questions in Qualtrics® they were returned to our platform, which automatically marked the daily diary as completed. This system allowed tracing of the number of responses by each participant without directly associating the participant’s identification data with the responses to the diaries. The answers of the initial measures and diaries were linked using a code only known by the participants.
Two months after collecting the diaries, a group of psychologist researchers returned to the centres that participated in the study. The objective was to collect data for the follow-up from participants who had previously collaborated. Specifically, participants answered the SA measure (Olivares et al., Reference Olivares, Ruiz, Hidalgo, García-López, Rosa and Piqueras2005).
Measures
Initial measure (person-level, Wave 1)
SA was measured with the Spanish version (Olivares et al., Reference Olivares, Ruiz, Hidalgo, García-López, Rosa and Piqueras2005) of the Social Anxiety Scale for Adolescents (SAS-A; La Greca and Lopez, Reference La Greca and Lopez1998). This scale consists of four distracting items (e.g. ‘I like to play sports’) and 18 items that measure SA (e.g. ‘I am ashamed to be surrounded by people I do not know’). The statements are rated on a 5–point Likert scale ranging from 1 (never) to 5 (all the time) in relation to the last month. The Cronbach’s alpha coefficient was .96 for the initial measure (Wave 1). This measure was also employed for the follow-up (Wave 3, Cronbach’s alpha = .96). The mean of items was used.
Interpretation bias in ambiguous social scenarios was measured with the Spanish version of the Adolescents’ Interpretation Bias Questionnaire 2.0 (AIBQ 2.0; Miers et al., Reference Miers, Sumter, Clark and Leigh2020; Spanish version: Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Miers and Calvete2022a). The AIBQ 2.0 describes three types of ambiguous situations: five non-social situations (e.g. ‘You have received bad marks for your last few tests. Why has this happened?’), five offline situations (e.g. ‘You have just given a presentation in front of your class and afterwards no-one asks a question. Why doesn’t anyone ask a question?’), and seven online situations (e.g. ‘You post a photo of a tasty dish that you have made on Instagram. After an hour, one of your followers responds, “What dish is that?”’). A question related to each situation is presented (e.g. ‘What is meant by this response?’) with a neutral interpretation (e.g. ‘It was nearly lunch, so everybody wanted to leave’), a negative interpretation (e.g. ‘They did not think my presentation was interesting’), and a positive one (e.g. ‘They thought what I said was very clear, and did not need to ask anything’). The participants were instructed to imagine the situation and to rate the probability that each interpretation would pop into their mind on a 5-point scale ranging from 1 (does not pop in my mind) to 5 (definitely pops up in my mind). Considering that interpretation bias is defined as a negative or threatening perception and only negative interpretations differentiate participants with low and high SA (Miers et al., Reference Miers, Blöte, Bögels and Westenberg2008), the data on negative interpretation were taken into account (Miers et al., Reference Miers, Blöte, Bögels and Westenberg2008). Because the online and offline dimensions of the scale are highly correlated, only an overall measure with these two components was used. Cronbach’s alpha coefficient was .89. The mean of items was used.
The Ambiguous Faces Interpretation Task was used to assess the interpretation bias of ambiguous faces (Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Mueller and Calvete2022b). This task consists of images of eight models (four men and four women) from the Chicago Face Database (Ma et al., Reference Ma, Correll and Wittenbrink2015). The task includes nine images of each model, so there are 72 images of real faces that express emotions of anger or happiness with different levels of ambiguity. Half of the images were created from the transformation of several intermediate images between a happy face and a neutral one. The other half combines intermediate images between an angry face and a neutral one (see section S1 in Supplementary material). The task consists of presenting a (+) signal for 500 ms followed by the presentation of one of the images (see section S2 in Supplementary material). The participants had to answer as quickly as possible whether the face seemed positive (pressing ‘P’ on the keyboard) or negative (pressing ‘N’ on the keyboard). Following the findings of Prieto-Fidalgo et al. (Reference Prieto-Fidalgo, Mueller and Calvete2022b), the index of the number of ambiguous faces marked as negative was used. The ambiguous faces are composed of the most ambiguous faces images; specifically, levels 4, 5 and 6 (see section S1 in Supplementary material; n = 24).
Daily assessment (day-level, Wave 2)
Social stressors were assessed using an ad hoc measure. Following the categories of social situations proposed by the Social Anxiety Questionnaire for Adults (CISO-A; Caballo et al., Reference Caballo, López-Gollonet, Salazar, Martínez Arias and Ramírez-Uclés2006), five everyday social stressors were defined: (a) public speaking or interaction with an authority (e.g. teacher); (b) sense of not having a sufficient ability to manage social situations; (c) having to express annoyance, anger or displeasure; (d) interaction with another person with sexual or romantic meaning; and (e) interaction with strangers. The participants were asked if any of the above-mentioned socially stressful situations had occurred throughout each day (e.g. ‘Have you had to speak or act in public or with any authority?’). In order to facilitate the understanding of the question, some examples were given for each social stressor (e.g. ‘Teacher asked me something in class, speaking in class or a meeting, speaking in public, talking to a teacher or superior’). In this study, the daily frequency of social stressors was considered.
To assess Safe-B, the items from the Spanish version of the Social Phobia Safety Behaviors Scale (SPSBS; Pinto-Gouveia et al., Reference Pinto-Gouveia, Cunha and do Cu Salvador2003; Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Miers and Calvete2022a) were adapted to a daily assessment methodology. The original version of SPSBS is composed of 17 statements. Because some of the items of the original version reflected similar reactions to a social situation, it was possible to construct a reduced list of 11 Safe-B. For example, items 4, 5, 6, 7 and 8, which refer to avoiding attracting attention, were included as a single item (‘Avoid being in a place or doing something that attracts attention’), and items 12, 13 and 16, which represent behaviours through which individuals try to mask their nervousness, were also grouped into one item (‘Trying to pretend that I’m comfortable’). The participants were asked about which Safe-B they engaged in.
Perceived SA was measured with a single item: ‘Indicate the degree to which today you have felt anxious in relationships with other people or in situations where you have been observed and/or evaluated (including situations through the internet or social networks)’. The response ranged from 0 (no anxiety) to 100 (very high anxiety).
Data analysis
The analysis was carried out with multi-level structural equation models (MSEM; Sadikaj et al., Reference Sadikaj, Wright, Dunkley, Zuroff, Moskowitz and Rauthmann2021). To test the hypotheses, two different models were estimated, with the only difference being that in the first model, the outcome was the daily SA (level 1; Fig. 1A), and in the second model the outcome was the level of SA at the 2–month follow-up (level 2; Fig. 1B).
At level 1 (day level), the daily SA predictor model included the association between daily stressors, daily Safe-B, and daily SA. At level 2 (person level), the model included the initial interpretation bias and initial SA measures predicting daily Safe-B and daily SA. In addition, the initial interpretation bias and initial SA were included as moderators of the relationship, on one hand, between stressors and Safe-B, and on the other hand, between stressors and SA. Moreover, the response day (time) was included at the day level to correct for the influence of time.
The predictor model for SA at follow-up (Fig. 1B) was similar to the model with SA in diaries, except that the outcome was SA assessed 2 months after completion of the diaries (level 2). For a similar procedure, see Seo et al. (Reference Seo, Lee, Jamieson, Reis, Josephs, Beevers and Yeager2022). The two models were performed with Mplus 8.2 (Muthén and Muthén, Reference Muthén and Muthén2021) using robust maximum likelihood estimation (MLR).
Results
Descriptive data and correlations between study variables are shown in Table 1. Higher levels of interpretation bias in ambiguous scenarios and with ambiguous faces were significantly and positively associated with higher levels of SA in the initial stage, in diaries and at follow-up. Safe-B in diaries were also significantly associated with higher interpretation bias in ambiguous scenarios, interpretation bias with ambiguous faces, daily SA, and initial SA levels. In addition, Safe-B in diaries were also associated with a higher level of SA at follow-up. The number of stressors was significantly and positively associated with the number of Safe-B performed and SA experienced on the same day. The time variable, which indicates the day on which the response is given, was only negatively and significantly associated with the number of Safe-B performed.
Descriptive analysis and correlation matrix between day-level and person-level variables
Note. SA = social anxiety; IB Scenarios = interpretation bias of ambiguous scenarios measured with AIBQ 2.0; IB Faces = interpretation of ambiguous faces measured with Ambiguous Faces Interpretation Task; Safe-B = safety behaviors; Stressors = number of social stressors experimented; W1 = wave 1 or person-level initial measure; W2 = daily level or diaries; W3 = measure at follow-up. *p < .05; **p < .01; ***p < .001.
Daily social anxiety model
The intraclass correlations of the daily SA model for Safe-B (ICC = 0.61) and SA (ICC = .59) showed sufficient intra- and inter-subject variability to justify the suitability of the day-level analysis.
In relation to the first objective, as shown in Fig. 2, the model results indicate that daily social stressors significantly and positively predicted daily Safe-B (H1) and daily SA (H6). In addition, initial interpretation bias in ambiguous scenarios predicts more Safe-B (H3) but not more daily SA (H8). Nevertheless, the interpretation bias measured with ambiguous faces did not predict more Safe-B (H2) or daily SA (H7). In relation to the mediational hypotheses, more initial interpretation bias in ambiguous social scenarios predicted a high frequency of Safe-B (H3), and a higher frequency of Safe-B was associated with a higher level of SA (H4). As expected, the indirect path between interpretation bias in ambiguous scenarios and SA via Safe-B was found to be statistically significant (section S3 in Supplementary material). These data supported the hypothesis of the mediational role of Safe-B between interpretation bias in ambiguous scenarios and experienced SA.
Path diagram of the model with social anxiety at the daily level. IB Scenarios, interpretation bias of ambiguous scenarios measured with AIBQ 2.0; IB Faces, interpretation of ambiguous faces measured with Ambiguous Faces Interpretation Task; Stressors, number of social stressors experimented; W1, wave 1 or person-level initial measures; W2, daily level or diaries. Only significant paths were represented. The dotted lines represent moderation paths. *p < .05; ***p < .001.
Regarding the second objective, according to the moderation of interpretation bias, none of the interpretation bias measures moderated the relationship between daily social stressors and Safe-B (H11 and H12). However, interpretation bias measured with ambiguous faces (H13) but not with ambiguous scenarios (H14) significantly moderates the relation between stressors and daily SA. Although the moderation was in the opposite direction of what was expected (Fig. 3A), the slope between the bias and SA was not significant when the participants experienced a high frequency (+1SD) of social stressors (B(134) = –0.36, t = –1.24, p = .22) or a low frequency (–1SD) of social stressors (B(134) = 0.06, t = 0.24, p = .81). Thus the data do not show a clear relationship between interpretation bias measured with ambiguous faces and SA. In sum, the data do not support the main hypothesis that bias would enhance the effect of stressors on Safe-B or daily SA. Rather, both social stressors and interpretation bias in ambiguous scenarios would independently explain the performance of Safe-B.
Moderation effects of model 1. A, moderation of the initial level of faces interpretation bias between social stressors and social anxiety at a daily level; B, moderation of the initial level of social anxiety between social stressors and safety behaviours at a daily level.
The third objective was to examine whether initial SA predicted Safe-B and whether SA moderated the relationship between social stressors and Safe-B. Initial SA significantly predicts daily Safe-B (H15) and daily SA (H16). Furthermore, initial SA moderates the relationship between daily stressors and Safe-B (H18) but not the relationship between daily stressors and daily SA (H19). Thus, the initial SA explained the number of Safe-B performed independently and in interaction with the stressors.
Figure 3B shows the form of this interaction. Specifically, SA in the initial stage was more associated with performing Safe-B when the participants experienced a high frequency (+1SD) of social stressors (B(134) = 0.63, t = 4.32, p < .001) compared with when they experienced a low frequency (–1SD) of social stressors (B(134) = 0.23, t = 21.30, p < .001). Additionally, Safe-B also mediated the associations between daily social stressors and initial SA, on the one hand, and daily SA, on the other hand (section S3 in Supplementary material).
Although interpretation bias with ambiguous faces did not predict Safe-B or SA levels, interpretation bias with ambiguous faces co-varied positively and significantly with interpretation bias using ambiguous scenarios and prior SA levels. Finally, the passage of days did not explain the variation in the number of Safe-B performed or the perceived SA. Therefore, as expected in a temporal record of 7 days, no evolution was found in the Safe-B or SA.
Follow-up social anxiety model
In the model with SA in the follow-up, the outcome of daily anxiety level was replaced by an SA measure assessed 2 months after collecting the diaries. Most of the results found in the previous model were systematically replicated in this one (Fig. 4). The main difference from the previous model was that in the present model only initial SA predicted the SA levels 2 months after collecting the data from the diaries (H17). Therefore, as Safe-B did not predict the SA levels at follow-up (H5), the indirect paths underlying the relationship of interpretation bias and SA through Safe-B were not statistically significant (section S4 in Supplementary materials).
Path diagram of the mediational model with social anxiety follow-up. IB Scenarios, interpretation bias of ambiguous scenarios measured with AIBQ 2.0; IB Faces, interpretation bias of ambiguous faces measured with Ambiguous Faces Interpretation Task; Stressors, number of social stressors experimented; W1, wave 1 or person-level initial measure; W2, daily level or diaries; W3, wave 3 or person-level follow-up. Only significant paths were represented. The dotted line represents moderation path. *p < .05; ***p < .001.
Discussion
SA constitutes a psychological problem of great relevance due to its magnitude and the negative consequences for the people who experience it. This study attempted to provide answers to some of the existing knowledge gaps regarding the mechanisms involved in daily SA experiences and focused on two vulnerability factors: Safe-B and interpretation bias. To examine the relationships between initial levels of interpretation bias, diary level of Safe-B, and diary level of SA, an intensive longitudinal design was utilized.
The first objective was to assess whether Safe-B mediated the association between interpretation bias and SA. As expected, daily social stressors and initial interpretation bias were associated with more Safe-B, and daily Safe-B were associated with higher SA levels. Thus, consistent with the proposed hypothesis, the results showed that the relationship between interpretation bias in ambiguous scenarios and daily SA was explained by Safe-B. This indirect relationship was found in a cross-sectional study when interpretation bias was measured in ambiguous scenarios (Prieto-Fidalgo et al., Reference Prieto-Fidalgo, Miers and Calvete2022a) and in a longitudinal study when it was measured with ambiguous faces (Prieto-Fidalgo and Calvete, Reference Prieto-Fidalgo and Calvete2023). In fact, the data are in line with Clark and Wells’ cognitive model for SA (Clark and Wells, Reference Clark and Wells1995; Leigh and Clark, Reference Leigh and Clark2018), which proposes that in the face of a social stressor, negative social cognitions lead to the performance of Safe-B, and the increase in Safe-B leads to an increase in the somatic and cognitive symptoms of SA. Although apparently this mediation has not been previously analysed with intensive longitudinal methodologies (diary assessment), the results are consistent with the above mentioned studies (Prieto-Fidalgo and Calvete, Reference Prieto-Fidalgo and Calvete2023) and with cross-sectional studies which find that both interpretation bias (Beard and Amir, Reference Beard and Amir2010; Chen et al., Reference Chen, Milne, Dayman and Kemps2019) and the performance of Safe-B are associated with experiencing higher levels of SA (Hoffart et al., Reference Hoffart, Borge, Sexton and Clark2009). In relation to the direct association between interpretation bias and SA, despite the literature showing that interpretation bias predicts SA directly (Chen et al., Reference Chen, Milne, Dayman and Kemps2019), the present data demonstrate that this relationship is only explained through Safe-B.
The results regarding the interpretation bias with ambiguous face measurements were not as expected. The task did not predict the number of Safe-B performed or the level of daily SA. These results are consistent with those obtained by Chen et al. (Reference Chen, Milne, Dayman and Kemps2019), who found that interpretation bias with ambiguous faces did not predict higher levels of SA. However, the data from this study indicate that this measure is cross-sectionally correlated with the interpretation bias in ambiguous scenarios and, as in other studies (Chen et al., Reference Chen, Short and Kemps2020; Maoz et al., Reference Maoz, Eldar, Stoddard, Pine, Leibenluft and Bar-Haim2016), with SA levels at follow-up. In interpreting the results, it must be considered that interpretation bias related to ambiguous faces, compared with interpretation bias in ambiguous scenarios, had a significantly weaker relationship with SA. This has been found in other studies, which despite arguing that the use of visual stimuli could be more ecological (Heuer et al., Reference Heuer, Lange, Isaac, Rinck and Becker2010), found that the association between interpretation bias and SA is greater when verbal stimuli are used (scenarios, words, sentences, or vignettes) than when visual stimuli are used (Chen et al., Reference Chen, Short and Kemps2020; Hirsch et al., Reference Hirsch, Meeten, Krahé and Reeder2016). Thus, coupled with the fact that the interpretation bias with ambiguous faces co–varied with the interpretation bias in ambiguous scenarios, the construct underlying the interpretation bias could have materialized through the interpretation bias in ambiguous scenarios.
In congruence with the theoretical models (Clark and Wells, Reference Clark and Wells1995; Leigh and Clark, Reference Leigh and Clark2018), the data also indicate that social stressors lead to the performance of Safe-B, which in turn results in higher levels of SA. In this case, daily stressors also directly predict SA. That is, the effect of daily stressors on daily SA can be explained directly and through Safe-B. Prior to this study, other investigations have found a direct relationship between daily stressors and SA (Auerbach et al., Reference Auerbach, Richardt, Kertz and Eberhart2012).
The results with respect to the second objective discussed above refer to the prediction of daily SA. Performing Safe-B during the days of the daily study did not predict SA levels 2 months after completing the diaries. In fact, the level of SA was only predicted by the initial level of SA itself. This idea contradicts the current literature, which maintains that both interpretation bias (Beard and Amir, Reference Beard and Amir2010) and Safe–B (Gangemi et al., Reference Gangemi, Mancini and van den Hout2012; Leigh et al., Reference Leigh, Chiu and Clark2021; Piccirillo et al., Reference Piccirillo, Taylor Dryman and Heimberg2016) are fundamental factors in the maintenance of SA. This may partly be due to the high stability level of SA found in the sample between initial measure and follow-up (r = .86). The low variation in SA levels would not allow the identification of other mechanisms that could explain the change in SA. Another possible cause of the differences between daily and follow-up SA models could be the use of different instruments – a single item for daily assessment and a validated psychometric instrument for the follow-up.
The second objective was to assess whether interpretation bias moderates the association between stressors and Safe-B and stressors and SA – daily and at a subsequent follow-up. Regarding Safe-B, neither interpretation bias in ambiguous scenarios nor interpretation bias with ambiguous faces moderated the association between social stressors and Safe-B. Regarding SA, although interpretation bias in ambiguous faces significantly moderated the association between social stressors and SA, the association between interpretation bias in ambiguous faces and SA was not significant in any case. Consequently, this result does not allow us to draw solid conclusions. Thus, future studies should study this effect in more detail. In addition, interpretation bias in ambiguous scenarios did not moderate the relationship between stressors and SA. The results align with previous studies examining the moderating role of other cognitive vulnerabilities for SA. For example, in the specific case of early maladaptive schemas, except for the schema of dependency, the results indicated that schemas did not moderate the effect of stressors on the prediction of SA (Calvete et al., Reference Calvete, Orue and Hankin2015; Cámara and Calvete, Reference Cámara and Calvete2012).
The third objective was to examine the influence of previous SA levels on the prediction of daily Safe-B and SA. Indeed, previous SA levels moderated the effect of daily stressors on Safe-B. The results showed that the initial SA level predicted greater use of Safe-B, especially with the co-occurrence of social stressors. Specifically, the relationship between stressors and Safe-B was stronger among participants with higher levels of SA. These data are consistent with other studies that have found that people with SA are more sensitive to social stressors. For example, in a study involving diaries, Farmer and Kashdan (Reference Farmer and Kashdan2015) found that the diagnosis of SA moderated the relationship between negative social events and the experience of negative emotions. Another study using an experimental methodology also showed greater sensitivity to social stressors in people with high levels of SA (Goldin et al., Reference Goldin, Manber, Hakimi, Canli and Gross2009).
Moreover, it was found that Safe-B mediated the described moderation relationship between initial SA and social stressors and the daily SA level. In this way, the current study contributes to revealing one of the mechanisms that could be involved and underlines the role of Safe-B in the maintenance of SA. Individuals with higher SA levels use more daily Safe-B, especially when social stressors occur. Furthermore, both stressors and Safe-B increase daily SA levels.
The results of this study must be interpreted in light of its limitations. First, only model 2 (Follow-Up SA Model) is fully longitudinal. Thus, the indirect relationship between interpretation bias and SA through Safe-B should be viewed with caution.
Second, as it was necessary for the diaries to be designed to be answered in a few minutes, SA was evaluated using a single item. Although numerous studies corroborate the validity of evaluation using a single item (Turon et al., Reference Turon, Carey, Boyes, Hobden, Dilworth and Sanson-Fisher2019), the measure’s reliability may have been compromised. Also, in relation to the diary assessment measures, specifically with the Safe-B measure, each Safe-B was only counted once, even if participants carry out more than once in a day. Third, similarly, the number of variables measured in the diaries was limited. According to the theoretical models, post- event processing is a relevant variable not included in the model that could influence the relationships analysed, especially the relationship with SA at follow-up (Blackie and Kocovski, Reference Blackie and Kocovski2018; Gavric et al., Reference Gavric, Moscovitch, Rowa and McCabe2017; Heimberg et al., Reference Heimberg, Brozovich, Rapee, Hofmann and DiBartolo2014). Thus, SA might increase only when the post-event processing tends to be negative. However, although more than 900 diaries were collected, the statistical power for person-level relationships (n = 116) is notably lower. Thus, the absence of significant prediction of SA at follow-up should be taken with caution. Fourth, although half of the sample reported SA levels comparable to the characteristics of clinical samples (Olivares et al., Reference Olivares, Ruiz, Hidalgo, García-López, Rosa and Piqueras2005), the sample does not necessarily represent a clinical sample. Future studies could increase the number of participants, use a clinical sample, improve the model by including other relevant variables, and use a multi-item measure to measure SA to overcome these limitations.
The data collection through diaries for 7 days and the longitudinal nature of the design enable conclusions to be drawn regarding the daily relationships between interpretation bias, stressors, Safe-B and SA. The results highlight the importance of Safe-B as a maintainer of daily SA. Its use not only contributes to people with high levels of SA experiencing anxiety in daily life, mainly when social stressors occur, but also mediates the effect of interpretation bias in ambiguous scenarios. These results may have clinical implications and suggest that interventions should include Safe-B reduction. This is consistent with the evidence regarding the efficacy of Safe-B-based interventions (Schmidt et al., Reference Schmidt, Buckner, Pusser, Woolaway-Bickel, Preston and Norr2012; Schreiber et al., Reference Schreiber, Heimlich, Schweitzer and Stangier2015). The results regarding interpretation bias also suggest that they should be modified, as has been addressed in various SA interventions (Naim et al., Reference Naim, Kivity, Bar-Haim and Huppert2018). Even so, without further studies to corroborate the results, the interpretation bias analysed does not seem to moderate the relationship between the experience of social stressors and Safe-B or SA symptoms.
Hence, social stressors and interpretation bias in ambiguous scenarios would independently lead to the performance of Safe-B.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1352465823000358
Data availability statement
The data that support the findings of this study are available from the corresponding author, A.P.-F., upon reasonable request.
Acknowledgements
None.
Author contributions
Ángel Prieto-Fidalgo: Conceptualization (equal), Data curation (lead), Formal analysis (lead), Funding acquisition (equal), Investigation (equal), Methodology (equal), Project administration (lead), Resources (equal), Software (equal), Supervision (supporting), Validation (supporting), Writing – original draft (lead), Writing – review & editing (supporting); Esther Calvete: Conceptualization (equal), Data curation (supporting), Formal analysis (supporting), Funding acquisition (equal), Investigation (equal), Methodology (equal), Project administration (equal), Resources (lead), Software (lead), Supervision (equal), Validation (equal), Visualization (equal), Writing – original draft (supporting), Writing – review & editing (lead).
Financial support
This research was supported by a grant from the Basque Government (PRE-2019-1-0034) to the first author.
Competing interests
We have no known competing interests to disclose.
Ethical standards
The research has conformed to the Declaration of Helsinki and the Ethics Committee of University of Deusto (reference no. ETK-4/20-21) approved the procedure of this study. For minors, an active consent from the parents was required to complete the daily assessment. All participants were informed of the tasks they were about to perform, and they consented to participate.
Comments
No Comments have been published for this article.