Test of common method bias

The Harman one-factor method was used to test for common method bias prior to data analysis. And the results revealed that there were 9 factors with eigenvalues greater than 1, as well as the variance explained by the first factor was 20.28%, which is far less than the critical value of 40% [28]. Thus, there is no significant common method bias in current study.

Data description

Table 1 shows the mean (M) and standard deviation (SD) of CERQ for all samples. Of the 2711 valid samples in this study, 1065 (39.28%) individuals reported engaging in one or more instances of NSSI and were considered to have been involved in NSSI behavior.

Figure 1 reveals the correlation between all the variables for all samples. Results showed that all the correlations among variables were at significant levels (r = − 0.29 ~ 0.82, p < 0.05) except for the association between positive refocusing and depression (r = 0.03, p = 0.12), refocus on planning and anxiety (r = − 0.02, p = 0.21), positive reappraisal and catastrophizing (r = 0.02, p = 0.28) and blaming others (r = 0.03, p = 0.09).

Fig. 1

Correlation coefficient of each variable (n = 2711)

Latent profile analysis

Table 2 presents the fit indices for solutions ranging from one to six profiles. The results indicate that the five-profile model is the most optimal solution based on a combination of model fit and parsimony. We evaluated model fit using several indices: Akaike information criterion (AIC), Bayesian information criterion (BIC), adjusted Bayesian information criterion (aBIC), Lo-Mendell-Rubin likelihood ratio test (LMRT), Entropy, and Bootstrap likelihood ratio test (BLRT). AIC, BIC, and aBIC are used to select the most parsimonious model by balancing model fit with model complexity. Lower values of these criteria indicate a better balance between fit and complexity, preventing overfitting.

Table 2 Fit statistics for the latent profile analysis (n = 2711)

LMRT and BLRT were used to compare the goodness-of-fit between models, with P < 0.05 indicating that a model with k profiles is significantly better than a model with k-1 profiles. The five-profile model shows lower AIC, BIC, and aBIC values compared to the four-profile model and also demonstrates a superior entropy value compared to the six-profile model, suggesting clearer classification.

Furthermore, the additional class in the six-profile model may result from splits driven by extreme responses from a small subset of participants, particularly on maladaptive CERS dimensions such as ‘Self-blame’ and ‘Rumination’. These extreme responses likely distort the classification process, introducing artificial splits that lack meaningful theoretical differentiation. This interpretation is further supported by the observation that these additional classes lack clear conceptual boundaries and exhibit substantial overlap with the classes identified in the five-profile model (Details of the six-profile model are provided in Table S3).

In conclusion, the five-profile model of CERS items provides not only the best fit according to statistical indices but also the most interpretable and theoretically coherent classification solution.

Figure 2 shows the scores of the five latent profiles of adolescents’ CERS. The naming of the five latent profiles was based on the differences in their scores on adaptive and maladaptive CERS. The five profiles comprised 19.73%, 24.68%, 32.25%, 14.42% and 8.82% of adolescents.

Fig. 2

Estimated mean values for latent profiles (n = 2711)

Table 3 shows the results of the one-way ANOVA, indicating statistically significant differences between the various latent profiles on adaptive and maladaptive CERS. Post hoc tests revealed that adolescents belonging to C1 had the lowest scores on both adaptive and maladaptive CERS. This suggests that the CERS of this group of adolescents is more rigid, and they can be classified as the ‘Rigid group’; Adolescents belonging to C2 have scores closer to the mean for adaptive CERS and lower scores for maladaptive CERS, and they can be classified as the ‘Moderate adaptive-low maladaptive group’; Adolescents belonging to C3 have higher scores than C1 for adaptive CERS and lower scores than C5 for maladaptive CERS, suggesting that the CERS of this group is more markedly maladaptive, which can be referred to as the ‘Maladaptive group’; C4 adolescents exhibit higher scores in adaptive CERS, while their maladaptive CERS scores are close to the average. This can be classified as a ‘High adaptive-moderate maladaptive group’; C5 adolescents exhibit the highest scores in both adaptive and maladaptive CERS, suggesting that the CERS of this type of adolescents is more sensitive, which can be classified as ‘Sensitive group’.

Table 3 Differences in CERS across the five latent profilesAssociation between demographic characteristics and the latent profiles

Taking the ‘Sensitive group’(C5) as reference, the ‘Rigid group’ (C1), ‘Moderate adaptive-low maladaptive group’(C2), ‘Maladaptive group’(C3) and the ‘High adaptive-moderate maladaptive group’ (C4) were compared. Odds ratio (OR) results showed that the latent profile of CERS was influenced by demographic factors such as age, gender, whether it was an only child, and whether it has left-behind experience(P < 0.05). Results of multiple logistic regression analysis indicate that age and gender were significant predictors of the latent profile of adolescents’ CERS (Table 4). Specifically, for each one-year increase in age, adolescents were 0.812 times more likely to belong to the C2 than to the C5 (OR = 0.812, 95% CI = 0.66–0.99, p < 0.05). The odds ratio (OR) for males belonging to the C2 compared to the C5 was 0.698 (95%CI = 0.52–0.94, p < 0.05), indicating a lower likelihood than females.

Table 4 Multinomial regression analysis of different classes with C5 as referenceAssociation between the negative emotions, NSSI and the latent profiles

The results of the ANOVA (Table 5) indicated statistically significant differences (p < 0.01) in the scores of three negative emotions (anxiety, depression, and stress) among adolescents with different latent profiles of CERS. Further multiple comparisons revealed that the trend of differences in anxiety and stress scores among adolescents across the five latent profiles was the same, with the C2 having significantly lower scores than the C1, and that there was no significant difference between the scores of these two groups and the scores of the other three groups; among the other three groups, the C4 had the lowest anxiety and stress scores, the C3 had higher scores, and the C5 had the highest scores. Meanwhile, depression scores varied significantly between the five groups. The most C5 was followed by the C3, then the C1 in third place, the C4 in fourth, and finally the C2 in last.

Table 5 Differences in negative emotions across the five latent profiles

The c2 test results (Table 5) indicated statistically significant differences (p < 0.01) in the detection rates of NSSI among adolescents with different latent profiles of CERS. The detection rate of NSSI was significantly lower in the C2 than in the C1. Additionally, it was also significantly lower in the C4 than in the C3 and C5.