The concept of self-compassion, i.e., an open and caring attitude toward oneself in challenging situations, has gained significant attention in psychological and psychotherapy research (Craig et al. 2020; Ferrari et al., 2019), as well as within counseling and in the general community (Halamová et al. 2021). In contrast to self-judgment and criticism (Neff, 2003a, b), self-compassion includes emotion regulation strategies for dealing with life´s adversities and helps individuals tolerate negative emotions and enhances their ability to self-soothe (Fong & Loi, 2016; Stutts et al. 2018), which can help reduce mental health problems such as stress, anxiety, and depression (Bui et al. 2021; Dev et al. 2020; Ferrari et al., 2019; MacBeth & Gumley, 2012; Stutts et al. 2018). Self-compassion has been described as a personality disposition, but is also a skill that can be developed, and different psychotherapy interventions have been created to help both community and clinical populations cultivate a compassionate attitude toward themselves (Ferrari et al., 2019; Gilbert, 2019, 2020). Compassion-focused therapy, which was specifically developed to decrease shame in patients with mental health problems, has been found to increase self-compassion (Craig et al. 2020), and to give clinical populations strength to stand up for themselves instead of feeling ashamed of one’s mental health problems (Bratt et al. 2019, 2020; Craig et al. 2020).
The most commonly used measure for assessing self-compassion is the Self-Compassion Scale (SCS; Neff, 2003b; Neff 2019), originally developed with a sample of U.S. college undergraduates, and has since been evaluated in numerous studies worldwide (Brenner et al. 2017; Halamová et al. 2021; Muris et al. 2018; Muris & Petrocchi, 2017; Neff, 2016; Neff et al. 2017, 2019). The SCS comprises 26 self-report items, enabling researchers to compute a total score representing overall self-compassion or scores for 6 subscales representing distinct components of self-compassion (Neff, 2003b). The 6 subscales encompass: self-kindness (treating oneself with kindness); self-judgment (treating oneself judgmentally); common humanity (recognizing that difficulties are part of the human experience); isolation (viewing difficulties as personal failures); mindfulness (enduring painful feelings in a mindful way); overidentification (getting caught up in a negative storyline about oneself) (Neff, 2016). Each pair of subscales represents opposing ends of a spectrum: self-kindness and self-judgment, common humanity and isolation, and mindfulness and overidentification. Together self-kindness, common humanity, and mindfulness represent a compassionate way of responding to difficulties, referred to as compassionate self-responding, whereas self-judgment, isolation, and overidentification reflect an uncompassionate way of responding to oneself, referred to as uncompassionate self-responding (Neff et al. 2019; Neff, 2022).
Given the widespread use of the SCS in research, it is essential to assess its psychometric properties, and several studies have demonstrated good reliability and validity across different populations (Halamová et al. 2021; Neff et al. 2017, 2019). However, the underlying factor structure of the SCS is still not fully understood. There has been controversy surrounding whether self-compassion should be measured as a single unitary construct using a total SCS score or separated into distinct components of compassionate and uncompassionate self-responding (Brenner et al. 2017; Muris et al. 2018; Neff, 2016, 2020, 2022). Neff et al. (2019) conducted a comprehensive investigation involving 20 samples from different countries and languages, comprising a total of 11,685 participants from diverse backgrounds including 10 community, 6 student, and 1 mixed community/student samples, both English and non-English-speaking. The investigation utilized Confirmatory Factor Analyses (CFA) and Exploratory Structural Equation Modeling (ESEM) to examine five models, including the 1-factor, 2-factor correlated, and 6-factor correlated models, as well as the single-bifactor (one general self-compassion factor and six ancillary factors), and the correlated two-bifactor models (two correlated general factors representing compassionate or uncompassionate self-responding each with three ancillary factors). The aggregated results indicated that the 1-factor and 2-factor correlated solutions of the SCS had poor fit, whereas the 6-factor correlated, single-bifactor, and correlated two-bifactor models showed good fit when using ESEM. However, Neff argued that the modest loadings for the two factors in the 2-bifactor model suggested that they had poor specification, and she therefore rejected this model despite the good overall fit. As for the CFA models, the correlated two-bifactor model showed good fit, but yielded model identification issues in 40% of the samples. The 6-factor correlated CFA model showed acceptable-to-good fit, whereas the remaining CFA models showed poor fit. The authors concluded that while the total SCS score or the 6 subscale scores can be used, it is not recommended to use separate scores for compassionate self-responding or uncompassionate self-responding.
In contrast to the view that self-compassion should be measured as a single unitary construct, some scholars have advocated for distinguishing between positive and negative self-responding (Brenner et al. 2017, 2018; Halamová et al. 2021; Montero-Marin et al. 2018; Muris et al. 2016; Muris & Petrocchi, 2017). Halamová et al. (2021) replicated the study by Neff et al. (2019), across 11 different populations and 10 different languages, five of which were the same as in Neff et al., 2019), and they concluded that the positive and negative components of the SCS should be distinguished. This was due to the superior fit of a two-tier model that includes two correlated general factors, one for compassionate self-responding and one for uncompassionate self-responding. They also suggested that it is possible that self-compassion does reflect a single unitary construct, but that the SCS may be unable to capture it reliably due to the high number of reversed-scaled items, which often cluster together into a separate factor. Montero-Marin et al. (2018) also raised concerns about measuring self-compassion as a single holistic construct, noting that including reversed items of uncompassionate self-responding, unintentionally introduced new dimensions.
Even though the SCS has been translated and validated in approximately 20 different countries, the Nordic countries have not been studied regarding the psychometric properties (Halamová et al., 2021). In the Neff et al. study (2019), one study from Norway was included, however this study did not explicitly explore the psychometric properties. For more information about other European countries see Halamová et al. (2021) and Neff et al., 2019. In Sweden the psychometric properties and validity of the short form of the SCS have been examined (Alfonsson et al., 2023; Bratt & Fagerström, 2020), but the full SCS remains relatively unexplored. In a study by Ängeby et al. (2022) involving 287 midwives who completed the full scale, results indicated a high prevalence of low levels of self-compassion. The researchers explored the factor structure through exploratory principal component analysis, suggesting a 2-factor solution, with some items showing weaker loadings (Ängeby et al., 2022). Both subscales demonstrated good internal consistency; Cronbach’s alpha values of 0.85 and 0.87 respectively. Additionally, another Swedish study (Johansson et al., 2022) reported self-compassion levels within the same range as those reported in the original study by Neff (2003b). Furthermore, Johansson et al. (2022) demonstrated that self-compassion levels increased after an internet-based compassion intervention for Swedish healthcare professionals. However, further studies are needed to explore self-compassion in other Swedish populations. In addition, since there is no validation of the full scale in Sweden, but it is used in research as well as in clinical practice, such as outcomes after compassion-focused interventions, it is important to determine whether the Self-Compassion Scale is useful in a Swedish context.
Research has shown that self-compassion is negatively related to perceived stress (Bui et al., 2021; Ferrari et al., 2019; Fisher & Pidgeon, 2018; MacBeth & Gumley, 2012; Stutts et al., 2018). However, the causal relationship between self-compassion and various types of well-being has not yet been established (Zessin et al., 2015). Bui et al. (2021) found that people who actively deal with their problems (proactive coping) tend to have higher self-compassion. Both proactive coping and self-compassion are linked to lower stress levels. However, when self-compassion was considered, the direct link between proactive coping and stress disappeared. This suggests that self-compassion could be a key in reducing stress, acting as a bridge between proactive coping and lower stress levels.
Similarly, previous research has shown a negative relationship between SCS and alcohol use, (Brooks et al., 2012; Phelps et al., 2018; Wisener & Khoury, 2020). Studies suggest that as self-compassion increases, alcohol use tends to decrease, likely because a reduction in negative feelings leads to less drinking (Brooks et al., 2012; Neff, 2003a; Neff et al., 2018). In other words, people who are more self-compassionate are less likely to use drugs and alcohol to cope. However, it is important to note that the causal links remain unresolved in the literature (e.g., people who reduce heavy drinking might become more self-compassionate as a result of improving their lives).
In the present study, we evaluated the psychometric properties of a Swedish translation of the SCS including its internal consistency, test-retest reliability, and the factor structure. We also assessed the concurrent validity of the SCS by correlating it with measures of perceived stress and alcohol consumption. This study aimed to evaluate the reliability and validity of the Swedish version of the SCS within a Swedish university student population. Specifically, we examined the internal consistency, test-retest reliability, and the factor structure of the instrument. The five models proposed by Neff et al. (2019) were assessed using CFA and ESEM analyses. Additionally, we investigated concurrent validity by examining whether self-compassion correlated with perceived stress and alcohol consumption.
Method
Participants
Data collection for Sample 1 spanned from 2018 to 2019, whereas Sample 2 were collected in 2023. Eligible participants were second-year students at Linnaeus University, Sweden, with no exclusion criteria applied. Second-year students were chosen as they were likely to have settled into the university environment. Fifteen out of 35 departments were randomly selected from Campus Växjö and Campus Kalmar by drawing lots. These departments represented diverse academic fields including economics, health and life sciences, art and humanities, social sciences, and technology. For robust psychometric analysis, we aimed to recruit at least 300 participants for factor analysis to ensure sufficient statistical power (Tabachnick & Fidell, 2007). For the correlational analyses (Sample 2), we collected data during one full semester with the aim of recruiting at least 100 participants, which would yield over 80% power to detect medium-sized correlations (Faul et al., 2009).
Sample 1 consisted of 305 students: 209 (68.5%) women, 90 (29.5%) men, 3 (1%) other gender identity, and 3 individuals (1%) who did not report their gender. This gender distribution is similar to the Swedish student population (Dryler et al., 2016). The mean age was 25.6 years (SD = 6.4) for the female students, 25.5 years (SD = 6.8) for the male students, and 27.3 years (SD = 0.3) for students of other gender identities.
Sample 2 included a new group of 161 participants who completed the initial test. Of these, 106 female (66%) and 52 male students (32%), and 3 students (2%) of other gender identities. Fifty-two participants completed the re-test, comprising 39 (75%) women, 10 (19%) men and 3 (6%) other gender identities. The mean age for these re-test participants was 23.5 years (SD = 5.8) for female students, 24.5 years (SD = 6.2) for male students, and 24.0 years (SD = 2.3) for students of other gender identities.
Procedure
Sample 1: One week before data collection, students received oral and written information about the study from their teachers. After providing informed consent, students completed the SCS, PSS-14, and AUDIT in a fixed order within the classroom (taking approximately 10–20 min), while researchers waited outside.
Sample 2: For test-retest-reliability assessment, we recruited a new group of students from the same 15 departments via email. After obtaining consent, we distributed digital questionnaires. The students created an ID code to use in the follow-up assessment. Two weeks after their initial admission the students were invited to complete the same questionnaire for the retest.
Measures
Self-Compassion Scale (SCS)
The SCS is a 26-item self-report scale including 6 subscales: Self-kindness (5 items), Self-judgment (5 items), Common humanity (4 items), Isolation (4 items), Mindfulness (4 items), and Over-identified (4 items). Self-judgment, Isolation, and Over-identified are reverse scored, to produce an averaged total score of all 26 items. A sample item is “I try to be understanding and patient toward those aspects of my personality I don’t like”, answered on a scale from 1 (almost never) to 5 (almost always). Total scores range from 1 to 5. Cronbach’s alpha was 0.91 for both Sample 1 and Sample 2 in the present study (computed separately), which is comparable to the original study (Neff, 2003b), which reported a Cronbach’s alpha of 0.92.
The SCS was translated from American English to Swedish using a forward-backward translation process (Brislin, 1986). Two bilingual native English-speaking individuals with excellent knowledge of the field conducted the translations separately. In the last step, the English translation was retranslated to Swedish and the first and third authors checked the three translations’ correspondence to each other to identify possible differences in meaning between the forward-backward translations.
Perceived Stress Scale (PSS-14)
This self-report instrument includes questions to what extent the participant has perceived their situation as unpredictable, uncontrollable, and overwhelming during the previous month. Consisting of seven negatively worded and seven positively worded statements, for example ”In the last month, how often have you felt confident about your ability to handle your personal problems?” with each rated on a five-point scale ranging from 0 (never) to 4 (very often), with a higher total score indicating greater perceived stress (Cohen et al., 1983). The total score can range from 0 to 56 points. The Swedish version has shown good internal consistency, Cronbach’s alpha 0.82 (Eskin & Parr, 1996), and 0.84 (Eklund et al., 2014). Eklund et al. also found that the PSS-14 has good concurrent validity. In this study Cronbach’s alpha was 0.87.
The Alcohol Use Disorder Identification Test (AUDIT)
This self-report instrument includes 10 items measuring alcohol consumption, drinking behavior, and alcohol-related problems with scores between 0 and 4, and a maximum score of 40 (Babor et al., 2001; Saunders et al., 1993). An example is “How many drinks containing alcohol do you have on a typical day when you are drinking?” (Score 0) 1 or 2 (Score 1) 3 or 4 (Score 2) 5 or 6 (Score 3) 7, 8, or 9 (Score 4) 10 or more (Babor et al., 2001). The Swedish version of AUDIT has been shown to have satisfactory internal and test–retest reliability with Cronbach alpha coefficient 0.82 (Bergman & Källmén, 2002). In this study Cronbach alpha was 0.83.
Data Analyses
Confirmatory Factor Analyses (CFA) and Exploratory Structural Equation Modeling (ESEM) were performed on the Self-Compassion Scale using Mplus version 8.5. (Muthén & Muthén, 2023). To replicate previous analyses (Neff et al., 2019) we used weighted least squares mean- and variance-adjusted estimator (WLSMV), reverse-coded negatively keyed items, and employed the same five models that they examined with CFA and ESEM. In the CFA models, cross-loadings were constrained to zero so that items were only allowed to load on their respective a priori target factors. In the ESEM models, cross-loadings were not constrained to zero so that items were allowed to load on other factors than their a priori target factors. For all analyses, we followed their shared Mplus code (Neff et al., 2019).
Model 1 was a one-factor model in which all items were constrained to load on a single unitary self-compassion factor. Model 2 was a 2-factor model with correlated factors representing compassionate self-responding and uncompassionate self-responding. Model 3 was a six-factor model with six correlated factors of self-compassion which are the 6 subscales: Self-Kindness, Self-Judgment, Common Humanity, Isolation, Mindfulness, Overidentification. Model 4 was a single bifactor model with a general factor (self-compassion) and six ancillary factors (Self-Kindness, Self-Judgment, Common Humanity, Isolation, Mindfulness, Overidentification). Model 5 was a two-bifactor model including two correlated general factors (compassionate self-responding and uncompassionate self-responding) each with three ancillary factors. We used the same fit indices and thresholds as Neff et al. (2019) to interpret model fit: the comparative fit index (CFI; 0.90 acceptable, 0.95 good), the Tucker-Lewis index (TLI; 0.90 acceptable, 95 good), and the root mean-square error of approximation (RMSEA, 0.08 acceptable, 0.06 good).
The remaining analyses were conducted using Jamovi version 2.3.28. For assessing internal consistency, we computed both Cronbach’s α and McDonald’s ω. The latter is a more robust statistic that does not assume that items have identical loadings. For assessing test-retest reliability and associations with perceived stress and alcohol consumption, we computed Pearson product-moment correlation coefficients. For inferential statistics, we set the significance threshold to 0.05, two-tailed. We treated the aggregated scales as continuous, and scores that were at least 3 SDs above the mean were considered potential outliers. The AUDIT scales had four standardized scores between 3.06 and 3.71, indicating potential outliers. In contrast, the PSS-14 and the Self-Compassion Scales had no standardized scores greater than 3. Consequently, we report the correlation between AUDIT and self-compassion with and without the inclusion of these outliers. We did not transform any of the variables.
Results
The overall results of the factor analyses were similar to those reported by Neff et al. (2019), with the 1-factor model (with a unitary self-compassion dimension) and the 2-factor correlated model (with two unitary factors representing compassionate self-responding and uncompassionate self-responding) showing a poor fit to the data (Table 1). The 6-factor correlated CFA model indicated acceptable fit, whereas the bifactor CFA models did not converge, which was also the case in some of the datasets reviewed by Neff et al. (2019). The 6-factor, single bifactor, and two-bifactor ESEM models showed good fit to the data. In the two-bifactor ESEM models, the average standardized loading was 0.25 for the compassionate self-responding (loadings ranging from − 0.04 to 0.51, 6 out of 13 loadings were significant) and 0.23 for uncompassionate self-responding (loadings ranging from − 0.06 to 0.51, 8 out of 13 loadings were significant), which is comparable to the numbers reported by Neff et al. (2019). In the single bifactor ESEM model the average standardized loading for the general self-compassion factor was 0.56 (item loadings ranging from 0.19 to 0.73, all 25 loadings were significant). The 6 specific factors of this model exhibited average loadings (and ranges) of 0.33 (0.25–0.43) for self-kindness, of 0.21 (−0.15–0.47) for self-judgment, of 0.51 (0.21–0.77) for common humanity, of 0.43 (0.14–0.67) for isolation, of 0.36 (0.19–0.43) for mindfulness, and of 0.42 (0.19–0.74) for overidentification. These numbers are also similar to those reported by Neff et al. (2019) with self-judgment having the lowest average loading.
Table 1
Fit Indices for Confirmatory Factor Analyses and Exploratory Structural Equation Modeling Analyses (n = 305)
Model | CFA | ESEM | ||||
|---|---|---|---|---|---|---|
RMSEA, 90% CI | CFI | TLI | RMSEA, 90% CI | CFI | TLI | |
1-factor a | 0.124 [0.118, 0.130] | 0.773 | 0.753 | 0.124 [0.118, 0.130] | 0.773 | 0.753 |
2-factor b | 0.092 [0.087, 0.098] | 0.874 | 0.863 | 0.092 [0.085, 0.098] | 0.887 | 0.866 |
6-factor c | 0.078 [0.071, 0.084] | 0.916 | 0.903 | 0.041 [0.031, 0.051] | 0.984 | 0.972 |
single bifactor d | N/A | N/A | N/A | 0.037 [0.025, 0.047] | 0.989 | 0.978 |
two bifactor e | N/A | N/A | N/A | 0.027 [0.009, 0.039] | 0.994 | 0.988 |
Table 2 shows the descriptive statistics of all instruments used in Sample 1 and Sample 2. Perceived stress levels (Sample 1) were relatively high with a mean score of 26.08. For comparison, the cutoff for high levels of stress is a mean of 25 or higher for Swedish adults (Eskin & Parr, 1996). Self-compassion was within the normal range for both samples, both for the total scale and the various subscales, compared to the original study by Neff (2003b). Alcohol consumption (AUDIT) were relatively low with a mean score of 5.80, where the cutoff for hazardous drinking is a score of eight or higher (Bergman & Källmén, 2002). All scales and subscales demonstrated good internal consistency (reliability coefficients above 0.70), except for the mindfulness subscale which showed modest internal consistency (reliability coefficients ranging from 0.58 to 0.63 across the samples). Skewness and kurtosis values were generally close to zero, indicating approximately normal distributions, although AUDIT showed moderate positive skewness and kurtosis. There were three missing scores on the PSS-14, and three missing scores on the AUDIT.
Table 2
Descriptive summary of Sample 1 and Sample 2 (Time 1)
Sample | Variable |
n
| M
| SD
| skewness | kurtosis | α
| ω
|
|---|---|---|---|---|---|---|---|---|
1 | PSS-14 | 301 | 26.08 | 8.64 | 0.26 | −0.20 | 0.85 | 0.87 |
AUDIT | 302 | 5.80 | 4.63 | 1.27 | 1.52 | 0.82 | 0.83 | |
SCS Total | 305 | 3.08 | 0.65 | −0.11 | −0.48 | 0.91 | 0.92 | |
Self-kindness | 305 | 3.08 | 0.81 | 0.00 | −0.17 | 0.82 | 0.82 | |
Common Humanity | 305 | 3.14 | 0.88 | −0.04 | −0.47 | 0.76 | 0.77 | |
Mindfulness | 305 | 3.14 | 0.74 | 0.02 | −0.28 | 0.61 | 0.63 | |
Self-judgment | 305 | 3.05 | 0.82 | 0.10 | −0.61 | 0.76 | 0.77 | |
Isolation | 305 | 3.14 | 0.95 | 0.04 | −0.82 | 0.78 | 0.79 | |
Overidentification | 305 | 2.94 | 0.98 | 0.04 | −0.86 | 0.78 | 0.78 | |
2 | SCS Total | 161 | 2.84 | 0.66 | 0.35 | −0.10 | 0.91 | 0.91 |
Self-kindness | 161 | 2.91 | 0.83 | 0.37 | −0.34 | 0.79 | 0.80 | |
Common Humanity | 161 | 2.78 | 0.89 | 0.33 | −0.45 | 0.72 | 0.73 | |
Mindfulness | 161 | 3.08 | 0.76 | 0.15 | −0.17 | 0.58 | 0.62 | |
Self-judgment | 161 | 3.36 | 0.81 | −0.29 | −0.06 | 0.74 | 0.76 | |
Isolation | 161 | 3.03 | 1.05 | −0.04 | −0.90 | 0.80 | 0.81 | |
Overidentification | 161 | 3.31 | 0.98 | −0.30 | −0.70 | 0.76 | 0.76 |
Correlations between Self-Compassion Scale and alcohol consumption (the AUDIT) and perceived stress (the PSS-14) are shown in Table 3. As expected, the SCS correlated negatively with perceived stress, r(299) = −0.66, 95% CI [−0.72, −0.59], p < 0.001. This means that individuals who reported higher self-compassion on average reported lower perceived stress. There was also a small negative correlation between self-compassion total scale and AUDIT, r(300) = −0.16, 95% CI [−0.27., −0.05], p = 0.005. If the four participants with AUDIT scores 3 SDs above the mean were excluded, the correlation between AUDIT and self-compassion remained significant, r(296) = −0.14, 95% CI [−0.25., −0.02], p = 0.017. This result indicates that individuals who reported higher self-compassion on average reported lower alcohol consumption.
Table 3
Correlations between AUDIT, PSS-14, and SCS scales (and subscales) in Sample 1
Variable | PSS-14 | Audit | SCS | SK | CH | Mi | SJ | Is |
|---|---|---|---|---|---|---|---|---|
PSS-14 | — | |||||||
AUDIT | 0.13* | — | ||||||
SCS | −0.66*** | −0.16** | — | |||||
Self-kindness | −0.53*** | −0.17** | 0.83*** | — | ||||
Common Humanity | −0.38*** | −0.11 | 0.68*** | 0.59*** | — | |||
Mindfulness | −0.41*** | −0.13* | 0.70*** | 0.60*** | 0.59*** | — | ||
Self-judgment | −0.52*** | −0.11 | 0.77*** | 0.61*** | 0.28*** | 0.30*** | — | |
Isolation | −0.55*** | −0.12* | 0.78*** | 0.48*** | 0.35*** | 0.42*** | 0.60*** | — |
Overidentification | −0.57*** | −0.10 | 0.77*** | 0.46*** | 0.34*** | 0.41*** | 0.61*** | 0.64*** |
Out of the 161 participants completing the SCS at T1, 67 (42%) completed the SCS at T2. Due to a technical error, we could not match the data for 15 of these participants, resulting in 52 participants with test-retest data. Pearson correlation coefficients between T1 and T2 were significant for the total scale and all the 6 subscales (p < 0.001). Specifically, the correlations were very large for the self-compassion total scale, r(50) = 0.93, 95% CI [0.88, 0.96], self-kindness, r(50) = 0.81, 95% CI [0.69, 0.89], common humanity, r(50) = 0.78, 95% CI [0.64, 0.87], mindfulness, r(50) = 0.77, 95% CI [0.63, 0.86], self-judgment, r(50) = 0.86, 95% CI [0.77, 0.92], isolation, r(50) = 0.86, 95% CI [0.76, 0.92], and overidentification r(50) = 0.80, 95% CI [0.67, 0.88].
Discussion
We evaluated the SCS in a Swedish context. The psychometric qualities of the Swedish translation were generally good, using the fit indices provided by Neff et al. (2019). The Swedish version demonstrated similar properties to the original version, including good fit between the data and several of the hypothesized factor structures, as well as high internal consistency and test-retest reliability in a sample of Swedish university students. We examined five models with CFA and ESEM (1-factor, 2-factor correlated, 6-factor correlated, 1-bifactor, and 2-correlated-bifactors). The factor analytic results closely align with those reported by Neff et al. (2019). The 1-factor and 2-factor models fitted poorly to the data according to both the CFA and ESEM analyses. The ESEM 6-factor correlated, 1-bifactor, and 2-correlated-bifactors models showed good fit, whereas the CFA 6-factor correlated model showed acceptable fit. The CFA 1-bifactor and 2-correlated-bifactors models failed to converge, which was also the case with some of the datasets reviewed by Neff et al. (2019). Furthermore, as expected self-compassion correlated negatively with perceived stress and alcohol consumption/risk behaviors, which supports the concurrent validity of the scale.
Despite the extensive translation and validation of the Self-Compassion Scale (SCS) in different countries around the world, there is a notable gap in research within the Nordic context regarding its psychometric properties (Halamová et al., 2021; Neff et al., 2019). Our findings support the generalizability of the SCS to measure self-compassion in accordance with Neff et al. (2019). Neff et al. (2019) argued that the 1-bifactor ESEM model has the highest level of theoretical conformity with Neff’s original model (2003b), specifying that self-compassion is comprised of six components that interact as a global system. Regarding the 6-factor correlated models (six distinct factors related to self-compassion), our findings showed that the CFA model had an acceptable fit to the data, while the ESEM model indicated a good fit. This may have significant implications for both research and clinical practice in Sweden, as it validates the use of the SCS for assessing overall self-compassion as well as its individual dimensions (Self-kindness, Common Humanity, Mindfulness, Self-judgment, Isolation, Overidentification). The results in our study reveal that the total scale and all subscales exhibited good internal consistency (except for the mindfulness, which had modest internal consistency) and high test-retest reliability across two weeks. This indicates that it is possible to use the subscales independently, unlike in the short version SCS-SF, where almost all subscales exhibit low internal consistency (Bratt & Fagerström, 2020). Future studies may use a larger test-retest sample to be able to examine longitudinal measurement invariance of the Swedish translation to verify that the scale has a similar factor structure across repeated measurements (e.g., in a randomized controlled trial).
Consistent with expectations, the SCS exhibited a large negative correlation with a standard measure of perceived stress (the PSS-14), a result that supports the concurrent validity of the SCS scale. The result is in line with other studies showing that individuals who report higher self-compassion also has lower levels of perceived stress (Bui et al., 2021). Furthermore, a modest negative correlation was observed between SCS and AUDIT. Although the result is too weak to draw any conclusive inferences, it still indicates the expected direction that individuals who report higher self-compassion also have lower alcohol consumption, consistent with previous studies (Brooks et al., 2012; Phelps et al., 2018; Wisener & Khoury, 2020). In a study by Spillane et al. (2022), high levels of self-compassion in first nation adolescents were associated with lower risk of alcohol-related problems and less alcohol consumption (Spillane et al., 2022). Increasing self-compassion has been found to help university students to develop emotion regulation and self-care strategies (Dundas et al., 2017), and compassion interventions for students can mitigate levels of stress (Fong & Loi, 2016) as well as protect against harmful alcohol consumption (Berg et al., 2024). Although the reasons for the small negative correlation between self-compassion and alcohol consumption are not fully understood, several factors could explain this. Firstly, the low average AUDIT score of 5.80 (with a standard deviation of 4.63) among the students indicates that most respondents had relatively low levels of alcohol consumption, which could limit the variability needed to detect a stronger correlation. Secondly, alcohol use and its correlation with self-compassion might differ depending on the population. For example, in a student population, taking a day off to party instead of studying might be positively correlated with self-compassion (i.e., taking a break), which could attenuate the correlation. Finally, the lack of strong evidence regarding the causal direction of the association between alcohol use and self-compassion makes it difficult to predict the size of the correlation.
Limitations and Future Research
This study represents the first examination of the psychometric properties of the full Self-Compassion Scale (SCS) in a Swedish context, including an evaluation of the factor structure, internal consistency, test-retest reliability, and concurrent validity. The results suggest that it is possible to use the SCS subscales independently, unlike the short version (SCS-SF), where most subscales exhibit low internal consistency. The study demonstrates good concurrent validity by showing a negative correlation between the Self-Compassion Scale (SCS) and perceived stress.
Some limitations are worth mentioning. Firstly, the bifactor CFA models did not converge, potentially limiting our understanding of the scale’s underlying structure. Notably, the 2-correlated bifactor CFA model has shown identification issues in several countries, which may be due to overparameterization (Neff et al., 2019). Secondly, the mindfulness subscale exhibited low internal consistency, although the test-retest reliability was high. Given the small size of our test-retest sample and the impact of attrition, it is important to exercise caution when interpreting the coefficient estimates. Additionally, although a modest negative correlation exists between the SCS and AUDIT (alcohol consumption), it is insufficient to draw definitive conclusions. The AUDIT scores were positively skewed meaning that most participants in our undergraduate sample reported lower alcohol consumption. Further investigation with a broader sample is needed to understand this relationship fully.
In summary, our evaluation of the Self-Compassion Scale (SCS) in a Swedish context revealed generally good psychometric properties for the Swedish translation, closely resembling the original version. The factor analytic results aligned with Neff’s previous findings. Notably, the 1-factor and 2-factor models showed poor fit, while the ESEM 6-factor correlated, 1-bifactor, and 2-correlated bifactor models demonstrated good fit. Concurrent validity was supported by the negative correlation between self-compassion and perceived stress. However, the mindfulness subscale exhibited low internal consistency. Despite some limitations, we find that the SCS can be used in a Swedish context.
Acknowledgements
We want to thank Simon Lönn and Erik Wernstål for performing the test-re-test data collection.
Declarations
Ethics Approval
Ethical approval was obtained from the Regional Ethical Review Board in Linköping, Sweden (Number: 2018/314-31).
Informed Consent
Participants´ identities were not collected. Informed consent was obtained from all participants included in the study.
Conflict of Interest
The authors report no conflicts of interest.
Use of Artificial Intelligence
We have utilized the AI service Copilot to enhance the quality of the English language in our writing.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
