3.3 Moderators
In subsequent analyses, we tested the influence of different pre-pandemic risk factors (social anxiety, childhood trauma, life events) and resources (self-efficacy & coping strategies) on the trajectory of self-reported psychosocial strain.
Social Anxiety. Considering social anxiety as a risk factor, we found almost identical effects for the SPAI and LSAS, presumably due to the high correlation between questionnaires (r = .76, p < .001). Social anxiety showed a significant main effect (SPAI: F(1, 291) = 42.20, p < .001, ηp² = .13; LSAS: F(1, 291) = 53.44, p < .001, ηp² = .16), which denotes a positive correlation between social anxiety and strain at all time points (SPAI: rs ≥ .304; LSAS: rs ≥ .299). The interaction of social anxiety and time was also significant (SPAI: F(1.77, 515.44) = 12.02, p < .001, ηp² = .04; LSAS: F(1.78, 518.08) = 9.75, p < .001, ηp² = .03): The pre-pandemic strain was higher for participants who also reported stronger symptoms of social anxiety (SPAI: r = .68; LSAS: r = .67). Individuals with greater social anxiety, however, experienced a less pronounced rise in strain until the peak of the pandemic (SPAI: r = − .27; LSAS: r = − .26) followed by a decline to the relative downturn that was independent of social anxiety (SPAI: r = .00; LSAS: r = .03; see Fig. 3A and B). Only for the SPAI, we additionally observed a small but significant interaction with gender (F(1, 291) = 5.21, p = .023, ηp² = .02) that was driven by the correlation between SPAI and the average strain across all time points being higher for women (r = .59, p < .001) than for men (r = .38, p < .001). Other interactions did not reach statistical significance (SPAI: Fs ≥ 1.43, ps ≤ .233; LSAS: Fs ≥ 1.74, ps ≤ .189).
Self-Efficacy. For self-efficacy (GSE), similar results as for social anxiety were observed (see Fig. 3C). We found a main effect of GSE (F(1, 290) = 28.22, p < .001, ηp² = .09), reflecting an increase in strain with decreasing self-efficiency ratings across all time points (rs ≤ − .214). Additionally, an interaction of GSE and time was found (F(1.78, 515.76) = 9.89, p < .001, ηp² = .03). Pre-pandemic strain was greater for individuals with less self-efficacy (r = .56, p < .001) but they also experienced a smaller increase during the pandemic peak (r = .26, p < .001). The change from peak to downturn, however, was independent of self-efficacy (r = − .06, p = .261).
Emotion Regulation. Maladaptive emotion regulation strategies (CERQ-mal) showed the same pattern as the previous risk factors (Fig. 3D). There was a main effect of CERQ-mal on strain (F(1, 291) = 38.34, p < .001, ηp² = .12) that was reflected by positive associations between these two variables across all time points (rs ≥ .23). We also observed an interaction with time (F(1.80, 523.46) = 11.47, p < .001, ηp² = .04). While baseline strain was elevated for participants with maladaptive emotion regulation strategies (r = .62, p < .001), the rise during the first pandemic peak was less pronounced for these individuals (r = − .29, p < .001). The following decline until downturn was yet again independent of maladaptive emotion regulation strategies (r = .08, p = .145).
For adaptive emotion regulation strategies (CERQ-adapt), we found a small but significant main effect of time (F(1, 291) = 5.61, p = .019, ηp² = .02), which was due to participants with less elaborated adaptive emotional regulation strategies experiencing stronger psychosocial strain (rs ≤ − .05). Beyond this main effect, we could reveal a three-way interaction of CERQ-adapt, time, and lag (F(1.77, 516.14) = 3.41, p = .039, ηp² = .01), which in turn was superseded by a four-way interaction with gender (F(1.77, 516.14) = 3.26, p = .045, ηp² = .01). Clarifying the four-way interaction, further analyses revealed that the three-way interaction of CERQ-adapt, time, and lag was only significant for male (F(1.79, 123.75) = 3.38, p = .042, ηp² = .05) but not for female participants (F(1.76, 391.25) = 0.54, p = .560, ηp² < .01). As can be seen in Fig. 4A, men with elevated adaptive emotion regulation strategies were able to buffer against psychosocial strain during pandemic onset only if the lag between assessments was high (r = − .31, p = .006). The baseline difference in strain between males with low compared to high adaptive emotion regulation strategies did not significantly vary as a function of lag (r = − .18, p = .119).
Acceptance was treated as a separate predictor of the CERQ and did not show a significant main effect on psychosocial strain (F(1, 291) = 1.81, p = .180, ηp² < .01). However, a three-way interaction of acceptance, time, and gender emerged (F(1.77, 514.98) = 3.98, p = .024, ηp² = .01). As can be seen in Fig. 4B, only men benefited from acceptance, which buffered against the rise in strain that was observed in the whole sample during the first peak of the pandemic.
Childhood Trauma. Childhood trauma (CTQ) revealed similar effects as the risk factors described in Fig. 3. The main effect of the CTQ (F(1, 291) = 5.57, p = .019, ηp² = .02) denotes a generally positive association between childhood trauma severity and psychosocial strain but we also observed an interaction with time (F(1.79, 521.86) = 5.27, p = .007, ηp² = .02) that was driven by a baseline difference (r = .29, p < .001) followed by a reduced increase in individuals with higher CTQ (r = − .25, p < .001), resulting in similar strain for all participants during peak pandemic that was independent of childhood trauma (r = − .01, p = .913). The decrease in strain until pandemic downturn, however, was also smaller with increasing CTQ values (r = .12, p = .043) such that individuals showed small but significant differences in strain during the last assessment that could be predicted by childhood trauma severity (r = .12, p = .040).
Life Events. Prior experience of threatening events (LTE) had no modulatory effects on the group-level results reported in Fig. 2 (Fs ≤ 1.58, ps ≥ .210). Considering adverse life events (ALE), however, we found a five-way interaction of ALE × time × gender × age × lag (F(1.77, 515.81) = 3.22, p = .047, ηp² = .01). Individuals with an age below or equal to the median showed only a main effect of time (F(1.78, 317.12) = 12.14, p < .001, ηp² = .06) and a marginal main effect of life events (F(1, 178) = 2.84, p = .094, ηp² = .02). Participants with an age above the median exhibited a slightly more pronounced effect of time (F(1.75, 198.15) = 14.94, p < .001, ηp² = .12) but also an interaction of time and lag (F(1, 113) = 6.23, p = .014, ηp² = .05) and a marginal four-way interaction with time, lag, gender and ALE (F(1.75, 198.15) = 2.90, p = .064, ηp² = .03). Follow-up analyses indicated that only older males showed the interaction between ALE and lag (F(1, 40) = 5.30, p = .027, ηp² = .12) while older females did not (F(1, 73) = 1.65, p = .203, ηp² = .02). Finally, this interaction could be traced back to the ALE only showing a tendency for a main effect for older males with a low lag between first and last assessment (r = − .54, p = .086) but not when lag between first and last assessment was high (r = − .01, p = .964).