2.4. Statistical analysis

Mean (M) and standard deviation (SD) for continuous variables, and frequency and percentage for categorical variables were calculated via SPSS 23. Student’s t-test analyses were used to detect group differences (i.e. sex, having sibling[s] or not, residence, and family structure) of PTSD/DSO symptoms. Pearson correlations were conducted to examine associations between each two key continuous variables (SES, ACEs, PTSD, DSO, self-kindness, and self-judgement). Further statistical analyses were performed with structural equation modelling (SEM) in Mplus 8.0. To examine the mediating role of self-kindness and self-judgement in the relationship between ACEs and PTSD/DSO, a series of structural equation models were built. We first established a direct model from ACEs to PTSD and DSO, and relational paths among PTSD and DSO were added. In the second step, based on the direct model, a multiple indirect model with the mediators (self-kindness and self-judgement) was built inserted between adverse ACEs and PTSD/DSO. The specific mediation pathways were presented as below: ACEs → self-kindness → PTSD/DSO; ACEs → self-judgement → PTSD/DSO. The cumulative risk of ACEs has received greater attention in the present study, rather than the latent construct. Thus, ACEs were treated as observed variables with total scores used during data analysis (Bethell et al., 2017). Parcelling has certain advantages (e.g. reduction of random error), which may allow for a more precise interpretation of the relationships between the studied variables in SEM models (Little, Cunningham, Shahar, & Widaman, 2002; Matsunaga, 2008). Given that CPTSD included six factors/symptoms clusters (RE, AV, and TH for PTSD, while AD, NSC, and DR for DSO) (Ho et al., 2020), the CPTSD was categorized into six subscales, with each one including two items. Existing literature has not addressed regarding the unidimensionality (the prerequisite for parcelling) of SK and SJ, thus the original items of SK and SJ were retained in the SEM model. Despite little theoretical or empirical evidence supporting the possibility of a reverse mediation, we tested the reverse model (ACEs → PTSD/DSO → SK/SJ) statistically (Chang, Fehling, & Selby, 2020) because of a cross-sectional study. Of note, given that ACEs were retrospective events before 18 years old and current PTSD/DSO and self-kindness/self-judgement status might not have an influence on, models with ACEs as a dependent variable were not tested.

The analysis was performed using the robust weighted least squares estimator (WLSMV) for parameter estimation as it is suitable for categorical items (Muthén & Muthén, 2009). For all pathways, direct, specific indirect, total indirect (standardized and unstandardized effects), and total effects were estimated. Standardized values were reported for all estimations. The 95% bias-corrected bootstrap confidence intervals (CIs) of the indirect effect were calculated on the basis of 5000 bootstrap samples. Goodness of fit was assessed with the following fit indices: comparative fit index (CFI), root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR). Thresholds were used as follows: for CFI excellent fit > .95 and moderate fit > .90; for RMSEA and SRMR excellent fit < .05 and moderate fit < .08 (Hu & Bentler, 1999).