A central problem of network analysis of psychopathology is that either all items of a questionnaire or a selection of items only based on the items’ content is included in the network analysis. While the former approach can lead to a strong overlap between included components [
36], in the latter approach it is often not clear how the selection of items was achieved. Therefore, to prevent too much overlap between components and to base the item selection not only on theoretical but also on empirical criteria, we adopted an approach in which we empirically tested for the components and corresponding items of the SPAI-C we aimed to include in the current network. For this purpose, we conducted a confirmatory factor analysis (CFA) in which we assigned items of the SPAI-C to the seven components (symptoms and cognitive variables described in theory section—see also Table A in Supplements). In this context, the CFA functioned as an empirical tool. The various CFA fit indices indicate whether the seven components included in the network are empirically justifiable compared to factor solutions reported in the literature and the factor loadings show, which item is most representative of each component. For each of the seven components, we chose the item with the highest loading as representative to further reduce redundancy (see Table
2). Items that could not be assigned to a particular component (item 1, 10, 11, 12, 13, 18, 22, and 23) were excluded from the analysis. We then compared the fit indices of the CFA to test whether these seven resulting components prevailed over the factor solutions described in research (1-, 3-, 5-, and hierarchical 5-factor solution). For this purpose, various fit indices [
67] were used as measures of comparison (CFI, RMSEA, SRMR). Due to the ordinal scaling of the SPAI-C, we used the maximization likelihood method, which is robust to possible violations of the multivariate normal distribution. Accordingly, to calculate the factors already described in the literature, we used those items that were assigned to each of the described factors in previous studies [
55,
60‐
62]. According to all fit indices, our solution based on the seven components (χ
2 (116) = 222.722, p < 0.001; CFI = 0.965, RMSEA = 0.067, SRMR = 0.035) showed a descriptively better fit than the other factor solutions (5H factor solution (χ
2 (203) = 452.915, p < 0.001, CFI = 0.916, RMSEA = 0.078, SRMR = 0.048); 5-factor solution (χ
2 (198) = 434.193, p < 0.001, CFI = 921, RMSEA = 0.076, SRMR = 0.045); 3-factor solution (χ
2 (227) = 458.784, p < 0.001, CFI = 0.926, RMSEA = 0.071, SRMR = 0. 042); 1-factor solution (χ
2 (299) = 731.614, p < 0.001, CFI = 0.878, RMSEA = 0.084, SRMR = 0.049)). In addition, we compared the chi-square model fit between the model based on the seven components and the models reported in literature using a chi-square difference test. Our model showed significantly better fit compared to all others (
p < 0.001). Since the SPAI-C differentiates between physiological symptoms before (item 25) and in a social situation (item 26), we also checked whether an 8-factor solution with differentiated physiological symptoms adds value. Although fit indices were as good as for the solution based on the seven components, the goldbricker procedure [
68] showed that item 25 and item 26 were redundant. The goldbricker procedure checks whether two components show similar correlations with other variables [
47,
68]. Based on a minimum correlation of
r = 0.80 and a threshold of 0.25, it was indicated that item 25 and item 26 correlate higher than
r = 0.80 and have less than 25% different correlations, suggesting a huge thematic overlap. Therefore, we decided to include item 26 (physiological symptoms during a social situation
) as Clark and Wells’ (1995) model refers to physiological symptoms during a social situation. Concerning the resulting seven items, the goldbricker procedure did not indicate any redundancy. Descriptive statistics of all items and an overview of items assigned to components (symptoms and cognitive variables) derived from theory are reported in the supplements.
Table 2
Representative items of the 7 factor-solution
S3: Scared while doing something and being observed | Observational fear |
S5: Scared when answering questions in class or at group meetings | Performance fear |
S7: Scared to meet new kids | Interactional fear |
S21: Before going to a party, I think about what might go wrong | Negative expectation |
S24: When I am with other people, I think “scary” thoughts | Dysfunctional cognition |
S19: I avoid social situations | Avoidance behavior |
S26: When I am in a social situation, I feel (somatic symptoms) | Physiological symptoms |