This is a cross-sectional study in which a convenience sample of Italian high-school students interested in healthcare programs and undergraduate healthcare students at a north Italian university were invited to participate. Participation was encouraged with the chance of winning a €50 gift card, randomly awarded to one participant who completed the survey. Inclusion criteria were: (a) being 18 or older, (b) being native Italian speaker, and (c) signing the informed consent to participate. No exclusion criteria were applied.

Undergraduate healthcare students received an email invitation from the principal investigator, describing the study and providing the link to the survey, which required approximately 20 minutes. High school students aged ≥ 18 were invited to participate in three Open Days showcasing healthcare courses, which took place from December 2022 to March 2023.

This study was approved by the Ethics Committee of the University of Milano-Bicocca (No. 707, Protocol 0066628/2022) and complied with the Declaration of Helsinki. The SOCS adaptation to the Italian context followed standard forward and reverse translation procedures. Initially, three Italian clinical psychologists with good English proficiency provided a conceptual translation of the questionnaire items, discussing discrepancies and reaching a consensus on the item formulation. Then, an independent bilingual translator back-translated the resulting Italian version. Finally, an English native reviewed the two English versions of the SOCS to identify eventual incongruency in phrasing.

Data analysis was performed using SPSS v29 and Mplus v7.4. All variables were descriptively analyzed (mean, standard deviation, range, skewness, kurtosis, frequency, and percentages).

As a previous step to conducting the primary analysis of the present study, the factor structures of the Italian versions of the SOCS-O and SOCS-S were assessed. Confirmatory factor analyses (CFAs) with a diagonally weighted least squares (WLSMV) estimation method were conducted to assess dimensionality. Sample size adequacy for the CFA was evaluated based on a commonly accepted rule of thumb recommending a ratio of 10 participants for each item in the scale (Kline, 1994). The models proposed by Gu et al. (2020) were tested for both scales: all 20 items forming a single latent factor, five 5-item correlated factors, and five 5-item factors forming one higher-order latent factor. The following four fit indices were used to indicate model-data fit (Hu & Bentler, 1999): the Comparative Fit Index (CFI ≥ .90 for acceptable fit), the Tucker–Lewis Index (TLI ≥ .90 for acceptable fit), the Root-Mean-Square Error of Approximation (RMSEA ≤ .10 for acceptable fit) with a 90% confidence interval, and the Weighted Root-Mean-Square Residual (WRMR ≤ 1 for acceptable fit).

Cronbach’s α and McDonald's ω were used to investigate the internal consistency of the SOCS-O and the SOCS-S. Moreover, corrected item-total correlations (r_tot) were calculated for the SOCS-O and the SOCS-S items to examine how each item contributed to the overall scale. A coefficient lower than .30 indicates that an item is measuring something different from the scale as a whole (DeVellis, 1991). Intraclass correlation coefficients (ICC) were calculated based on a mean-rating (k = 2), absolute-agreement, 2-way mixed-effects model to evaluate the SOCS-O and SOCS-S test-retest reliability. Convergent validity was estimated by calculating Pearson correlation coefficients between the SOCS-O, the SOCS-S, and the remaining measures (SCS-SF, RSE, B-IRI, FCS, SWEMWBS, and BCSQ-12). The strength of the correlations was interpreted following Cohen’s guidelines (Cohen, 1988): small (r = .10–.29), medium (r = .30–.49), and large (r = .50–1.00).

As the primary analysis of the present study, we used path analysis models to explore the mediator role of self-compassion facets. This approach allows for simultaneous estimation of relationships among variables and includes mediation effects. In our investigation, we evaluated a model where self-esteem (RSE) served as the independent variable, the five self-compassion facets (SOCS-S) acted as mediators, and well-being (SWEMWBS) was the outcome. We calculated standardized regression coefficients (β) for bias-corrected bootstrapped indirect effects using 10,000 bootstrap samples, along with their standard errors and 95% confidence intervals (CI). Indirect effects were deemed statistically significant if the 95% CI did not encompass zero (Lockhart et al., 2011), and the model included compassion for others (SOCS-O) as a covariate. The dataset and syntax used for the present study can be found in Pérez-Aranda et al. (2025).

The survey was completed by 438 participants. However, 30 were excluded for two main reasons: failing to answer a control item designed to detect random responding (n = 15) or not being native Italian speakers (n = 15). Control items were embedded in some questionnaires to identify inattentive responses, while non-native speakers — mainly international students — were excluded to minimize potential cultural and linguistic bias. Therefore, 408 participants (84.6% women, with a M_age of 21.07) were included in the analyses. All data provided by the participants were complete; no missing responses were recorded. Most of them (n = 249, 61%) were university students of healthcare professions (e.g., medicine, nursing, midwifery), while a notable proportion were in their last year of high school (n = 155, 38%). One-third of the sample (n = 135) were working and, in terms of income, most considered their financial situation to be around average (see Table 1).

Based on the fit indices, significance of factor loadings, and previous validation studies, the five-factor hierarchical model was interpreted as best fitting the data both for the SOCS-O (CFI = .99, TLI = .99, RMSEA = .04; WRMR = 0.86) and the SOCS-S (CFI = .97, TLI = .96, RMSEA = .08; WRMR = 1.50). See the Supplementary Materials for more details on the descriptive statistics of the SOCS-O and the SOCS-S items, the standardized factor loadings for the hierarchical five-factor model, and the fit indices for all the tested CFA models.

Internal consistency of the SOCS-O and the SOCS-S factors was adequate, with Cronbach’s α and McDonald’s ω ranging from .67 to .93 (see Supplementary Materials). The r_tot of the SOCS-O and the SOCS-S items was greater than 0.30 in all cases, suggesting adequate homogeneity. Finally, test-retest reliability after one month was adequate, with ICC coefficients ranging from .73 to .86, which supports the temporal stability of SOCS scores.

Table 2 shows the correlation coefficients between scores on the SOCS-O, SOCS-S, and other constructs. The SOCS-O had significant and large positive correlations with the “empathic concern” and “perspective taking” subscales of the B-IRI, significant and small positive correlations with the SWEMWBS and the RSE, significant and small negative correlations with the “personal distress” subscale of the B-IRI, the FCO/S, and the “neglect” subscale of the BCSQ12; no significant correlations with the SCS-SF and the “lack of development” and “overload” subscales of the BCSQ12 were observed. The SOCS-S had significant and large positive correlations with the SCS-SF and SWEMWBS, significant and moderate negative correlations with the FCS and the “overload” subscale of the BCSQ12, significant and small positive correlations with the RSE and the “empathic concern” and “perspective taking” subscales of the B-IRI, significant and small negative correlations with the “personal distress” subscale of the B-IRI, the FCO, and the “neglect” subscale of the BCSQ12, and no significant correlation with the “lack of development” subscale of the BCSQ12. Pearson’s correlation coefficients among the SOCS subscale scores were also computed (supplementary materials).

The proposed model adequately fitted with the data (χ12 = 32.54, p < .001, CFI = .97, SRMR = 0.026). As depicted in Figure 1, the path analysis model showed that self-esteem significantly predicted recognizing suffering of oneself (β = 0.088, SE = 0.044; p = .048), feeling for the suffering of oneself (β = 0.252, SE = 0.050; p < .001), tolerating the uncomfortable feelings aroused in response to the suffering (β = 0.277, SE = 0.046; p < .001), and acting or being motivated to act to alleviate suffering (β = 0.231, SE = 0.048; p < .001). The associations between self-compassion facets were statistically significant (r ranging from .26 to .75; p < .001 in all cases). Conversely, self-esteem was not a significant predictor of understanding the universality of suffering (β = -0.026, SE = 0.047; p = .588).

In turn, well-being was significantly predicted by feeling for the suffering of oneself (β = 0.212, SE = 0.069; p = .002), tolerating the uncomfortable feelings aroused in response to the suffering (β = 0.273, SE = 0.059; p < .001), and acting or being motivated to act to alleviate suffering (β = 0.190, SE = 0.069; p = .006). On the contrary, well-being was not significantly explained by recognizing one’s suffering (β = -0.036, SE = 0.047; p = .450) or understanding the universality of suffering (β = 0.007, SE = 0.049; p = .883).

Finally, compassion for others (SOCS-O) was included as a covariate of the model, and its effect was significant in every variable (β ranging from 0.091 to 0.392; p < .05 in all cases). Overall, the associations proposed in the model explained 39% of the variance of well-being and between 10 and 15% of the variance of self-compassion facets. See R-square estimates in Figure 1.

Regarding the mediating effect of self-compassion facets, the path analysis model observed three significant indirect effects mediating the relationship between self-esteem (RSE) and well-being (SWEMWBS): the self-compassion facets of “feeling for the person’s suffering” (β = 0.054, 95% CI [0.011, 0.097]; p = .014), “tolerating uncomfortable feelings” (β = 0.075, 95% CI [0.034, 0.116]; p < .001), and “acting to alleviate suffering” (β = 0.044, 95% CI [0.005, 0.083]; p = .029). The remaining self-compassion facets, i.e., “recognizing suffering” (β = -0.003, 95% CI [-0.011, 0.005]; p = .430) and “understanding its universality” (β = 0.000, 95% CI [-0.002, 0.002]; p = .877), did not show a significant mediating effect.

Our findings align with previous research that observed a significant mediating or moderating effect of self-compassion on the relationship between self-esteem and well-being. However, by using the SOCS and the distinction between cognitive, affective, and behavioral facets of self-compassion, we suggest that these effects may be particularly associated with the affective and behavioral components of self-compassion. Our results indicate that a significant part of the well-being experienced by individuals with high self-esteem is due to their tendency to emotionally connect with their own suffering, to tolerate the uncomfortable feelings that may arise in response to that connection, and to take action to alleviate that suffering. The implications of these conclusions extend beyond the conceptualization of self-compassion; our findings can be directly applied to the psychotherapeutic context.