Psychometric Properties of the Slovenian Long and Short Version of the Self-Compassion Scale

The Self-Compassion Scale (SCS) is a widely used instrument for measuring selfcompassion. In this study we examined psychometric properties of the SCS long and short form in a Slovenian community sample (N = 442). The total score reliability was satisfactory both in the long form (α = .91) and short form (α = .81). The reliability of subscales in the long form was between .66 and .84, and relatively low in the short form (from .45 to .72). We performed a confirmatory factor analysis and examined a one factor, a two factor a six-factor correlated model, a higher order model, and a bi-factor model. The six-factor and bi-factor models showed the best fit for the SCS-LF, and the six-factor model was the only acceptable fit for the SCS-SF. The correlation between the long and the short form is very high (r = .96). The SCS correlates in a predictable way with satisfaction with life, well-being, and attachment styles. We concluded that Slovenian versions of both the long and the short form of the SCS have generally from acceptable to good psychometric properties. Results supported the use of both the total score and subscale scores of SCS-LF. The short version of the SCS can be used as a good and economical alternative when the aim is to use the total score.

PSIHOLOGIJA, 2019, Vol. 52(2), [107][108][109][110][111][112][113][114][115][116][117][118][119][120][121][122][123][124][125] There has been an increasing dialogue in recent years between Eastern philosophical thought and Western psychology (Rubin, 1996). Mindfulness has become one of the most important concepts in psychology and mental health in recent years, as evidenced by increasing number of peer-reviewed published research articles (Černetič, 2017;Fulton, 2014) and wide recognition of mindfulness-based interventions in mental health (Kachan, et al., 2017). Another important and increasingly researched concept is selfcompassion. Compassion is a well-known concept in our culture and is defined as sympathy and kindness towards those who are suffering, and being open to their pain instead of avoiding or disconnecting from it (Wispe, 1991). Selfcompassion is related to the concept of a more general type of compassion, except that it is oriented inwards. According to Neff (2003a), self-compassion "involves being touched by and open to one's own suffering, not avoiding it or disconnecting from it, generating the desire to alleviate one's suffering, and to heal oneself with kindness. Self-compassion also involves offering nonjudgmental understanding to one's pain, inadequacies, and failures, so that one's experience is seen as part of the larger human experience" (p. 87). Neff (2003aNeff ( , 2012) has developed a model of self-compassion based on three basic bipolar components: (1) Self-kindness-Self-judgment: Self-kindness refers to a warm, kind, and compassionate attitude towards oneself, especially in difficult times or when we experience negative emotions, feelings of inferiority or selfdoubt. It includes accepting imperfections and understanding that challenges or problems are a part of life. Self-kindness also means taking care of oneself in times of need, instead of being judgmental about negative thoughts and feelings. The opposite pole includes self-criticism and harsh judgment in times one feels down, which prevents the doing of what's necessary to help oneself; (2) Common humanity-Isolation: Common humanity means perceiving one's imperfections and problems as something most people experience at some point in life. It means that one realises that failures are a part of being human and that other people also struggle sometimes. On the other hand, isolation includes seeing personal problems as something only we struggle with, which creates feelings of separation and more isolation from others. In this case person feels desolate and alone instead of realising the experience is shared with others; (3) Mindfulness-Over-identification: Mindfulness is defined as a non-judgmental awareness of feelings and thoughts without trying to change or avoid them. A mindful person looks at them with curiosity and balanced awareness instead of overly identifying with them or ruminating about a negative internal state.
Based on three bipolar components, Neff constructed the Self-Compassion Scale (SCS; Neff, 2003b), which is the most commonly used self-reporting instrument for measuring self-compassion. The SCS consists of 26 items and measures six factors: Self-kindness, Self-judgment, Common humanity, Isolation, Mindfulness, and Over-identification. The scale also measures overall self-compassion. With the use of confirmatory factor analysis, Neff (2003b) found a good fit to a six-factor model, and to a single higher order factor. The scale was validated on an undergraduate student sample. Based on the findings of CFA, Neff (2003b) noted that both subscales and the total score could be used for measuring self-compassion.
The SCS has been increasingly used in studies, exploring the relations between self-compassion, psychological health and psychopathology. Self-compassion as measured by the SCS is negatively correlated with psychopathology, neuroticism, and negative affect (MacBeth & Gumley, 2012;. Positive relations were found between selfcompassion and happiness, emotional intelligence, positive affect, curiosity, optimism, extroversion, motivation, and constructive responses to stress (Breines et al., 2014;Pinto-Gouveia, Duarte, Matos, & Fraguas, 2013). Self-compassion is positively related to satisfaction with life and well-being (Neff, 2003b(Neff, , 2015Neff, Pisitsungkagarn, & Hsieh, 2015). Selfcompassion does not only decrease negative well-being components, but it also increases life satisfaction with hope having a mediating role (Yang, Zhang, & Kou, 2016).
People with mental illnesses tend to have lower scores on the SCS (Krieger, Altenstein, Baettig, Doerig, & Holtforth, 2013;MacBeth & Gumley;. According to Leary, Tate, Adams, Allen, and Hancock (2007), self-compassion moderates reactions to distressing situations and facilitates coping with negative events. In their research, they found that self-compassionate people judge themselves less harshly and that their selfevaluations and emotional reactions are more based on actual performance when compared to people with lower scores on self-compassion. People with higher self-compassion were also more willing to accept responsibility for their roles in negative events.
Development of self-compassion is related to early relationships with primary caregivers. Neff and McGehee (2010) found that higher self-compassion is associated with maternal support, while maternal criticism is connected to lower self-compassion. The degree of family functioning significantly predicted levels of self-compassion. Neff and McGehee (2010) note that warmness and supportiveness of parents may be reflected in more self-compassionate children's inner dialogues compared to cold and critical parents.
Self-compassion as measured by the SCS correlates with attachment styles (Neff & McGehee, 2010;Wei, Liao, Ku, & Shaffer, 2011). Significantly higher levels of self-compassion were found in adolescents and young adults with a secure attachment style (Neff & McGehee, 2010). Preoccupied and fearful attachment styles were associated with lower self-compassion. Pepping, Davis, O'Donovan, and Pal (2015) explain that capacity for self-compassion might be decreased in persons who have not experienced parental affection or responsive parenting. Higher attachment anxiety may reduce the capacity for self-soothing with self-compassion. Wei and colleagues (2011) also found a negative association between attachment anxiety and self-compassion. However they found mixed results regarding association between self-compassion and avoidant attachment.
The SCS is increasingly used in studies focused on exploring effects of mindfulness and compassion focused interventions on mental health. Research shows that mindfulness training is one important way to increase self-compassion (Neff & Germer, 2012). Besides mindfulness training, specific interventions focused on developing self-compassion have been developed, such as Mindful self-compassion (Neff & Germer, 2012) and Compassionate mind training (Gilbert & Procter, 2006). Participation in an eight-week Mindful selfcompassion training significantly increased self-compassion measured by the SCS compared to a waitlist control group (Neff & Germer, 2012).
Many studies across different countries found satisfactory reliability and validity of the SCS (Neff, 2015). Internal reliability has been confirmed as high in various studies and different populations. The SCS has good discriminant validity and is not significantly correlated with social desirability (Neff, 2003b). Multiple studies explored the generalisability of the factor structure across various countries and populations. The six-factor structure was confirmed on a population of students in Greece (Mantzios, Wilson, & Giannou, 2013), Turkey (Akin, Akin, & Abaci, 2007), Iran (Azizi, Mohammadkhani, Lotfi, & Bahramkhani, 2013), Germany (Hupfeld & Ruffieux, 2011), Italy (Petrocchi, Ottaviani, & Couyoumdjian, 2013), and Japan (Arimitsu, 2014). Portuguese research supported both a six-factor correlated model and a hierarchical model (higher-order factor of self-compassion) in clinical and nonclinical adult samples (Castilho, Pinto-Gouveia, & Duarte, 2015), and in a sample of adolescents (Cunha, Xavier, & Castilho, 2016). The hierarchical model was also confirmed in a sample of Chinese students (Chen, Yan, & Zhou, 2011).
Even though most of the research supported a six-factor structure, Lopez and colleagues (2015) conducted a large representative study from a community sample, and could not replicate the original factor structure. Exploratory factor analyses found two factors with good internal consistency, one consisting of negatively and the other of positively formulated items. These authors proposed that self-compassion and self-criticism (negative factor) should be considered different constructs. Neff (2015) argues that these findings could have been due to a method effect, as exploratory factor analysis is known to be sensitive to positive versus negative affect.
In recent years there has been controversy regarding the use of the total score of the SCS, as some studies did not support the hierarchical model. Williams, Dalgleish, Karl, and Kuyken (2014) examined the factor structure of SCS in three samples: a sample of adults, adults who practice meditation, and adults with diagnosed recurrent depression. In all three samples, the higher order hierarchical model did not have an acceptable fit. These authors proposed that the SCS seems to be better suited for measuring six facets of self-compassion rather than an overarching single factor of self-compassion.
Because of inconsistent findings regarding hierarchical model, Neff (2015) suggested that in the future research, exploration of a bi-factor model should be encouraged instead of the hierarchical model. In the bi-factor model each item loads on a general factor and also on a corresponding subscale factor. Regarding the SCS, a general factor relates to general self-compassion and group factors are six subscale factors. Neff (2015) notes that the bi-factor model is also theoretically consistent with her conceptualisation of self-compassion. In a recent study, Neff, Whittaker, and Karl (2017) researched the factor structure of the SCS using different factor models. Their study was done on four different samples: undergraduates, community adults, individuals practicing meditation, and a clinical sample. They have found that the six-factor correlated model showed the best fit across samples, while the one factor, the two factor, and higher order models had inadequate fit values. They have found that the bifactor model showed an acceptable fit in the student, community, and meditator samples. The authors conclude that the SCS can be used to measure both the six subscale elements of self-compassion and the overall self-compassion. Consistent with these findings is research done with a French version of the SCS, which found that the bi-factor model gives a good omega index suggesting the relevance of using the general self-compassion score along with six subscale factor scores (Kotsou & Leys, 2016). The omega index indicates the percentage of variance in the total score accounted for by both the total score and subscale factor scores.
Tóth-Király, Bőthe, and Orosz (2016) have found support for the bi-factor model, while employing exploratory structural equation modeling (ESEM) on a Hungarian sample. Cleare, Gumley, Cleare, and O'Connor (2017) found support for both bi-factor and six-factor structure of SCS on a UK sample. To conclude, research findings in the recent years are supporting both the six-factor structure and the bi-factor model. This means that SCS can be used for measuring both the six facets of compassion and general self-compassion. Raes, Pommier, Neff, and Gucht (2011) constructed and validated a shortform version of the Self-Compassion Scale (SCS-SF) with 12 items. The factor structure in the shortened scale was the same as in the long-form scale. Both have a higher order self-compassion factor, and six second-order factors. The correlation between the two forms was very high (r ≥ .97). Subscales of the SCS-SF have relatively low internal consistency (between .54 and .75 for the English SCS-SF) so they recommend using the overall self-compassion score instead of subscale scores. The authors concluded that the short version of SCS can be used as an economical alternative to the full version in research and clinical settings. Castilho and colleagues (2015) also found that the Portuguese short version of the SCS is a reliable instrument for the assessment of self-compassion in clinical practice. PSIHOLOGIJA, 2019, Vol. 52(2), 107-125 The aim of this study was to research the psychometric properties of the Slovenian version of the SCS. We examined the reliability, and the factor structure of both the long and the short version of the SCS. We also explored convergent validity of the SCS, focusing on correlations with attachment styles, satisfaction with life, and well-being. As has been already discussed, higher self-compassion is related to attachment security, higher satisfaction with life and well-being. We expected to find low to moderate significant correlations between these constructs and self-compassion as was found in previous studies (Neff, 2003b;Neff & McGehee, 2010;Neff, 2015;Neff et al., 2015;Wei et al., 2011;Yang et al., 2016).

Method Participants
Participants took an online survey that was posted and shared through social media and e-mails. Out of a larger number of participants who opened online questionnaire (N = 1383), 442 were included in the main analysis (125 men, 317 women; M age = 31.5, SD = 11.3). The rest of the participants were excluded because of partial answers, most of them quit before or after completing demographical questions. A majority of participants had a university degree or professional qualification (58%). Ten percent of participants had a Master's degree or PhD, 30% of participants completed high school or vocational education, and three percent completed primary school. Most participants were single (31.7%), or in a relationship but living separately (32.8%), and 19% of participants were married. Participants were not compensated for their cooperation.

Procedure
We used back-translation procedure to translate SCS to Slovenian language. A translator created a back-translated version, which was compared to the original version. Inconsistencies in back-translation were discussed and resolved by authors of the article. Data for this study were collected between December 2015 and August 2016 in two waves. After the first collection of data, we decided to repeat the procedure because we wanted to increase the percent of male participants in our sample. In both cases, we prepared a battery of online questionnaires using site 1ka. On the first page of the questionnaire, we described the general aims of the study, and emphasized the voluntary participation and confidentiality of answers. The second page contained several sociodemographic questions. Completing the questionnaire took 15 minutes on average.

SCS-Long Form (SCS-LF; Neff, 2003b).
The SCS-LF is a self-report questionnaire with 26-items that measure six components of Self-compassion. Three of these six components measure positive aspects: Self-kindness (e.g., "I try to be understanding and patient toward those aspects of my personality I don't like."); Common humanity (e.g., "I try to see my failings as part of the human condition."), and Mindfulness (e.g., "When something painful happens I try to take a balanced view of the situation."). Self-kindness scale consists of five items, while Common humanity and Mindfulness scales consist of four items each. Other three components measure negative aspects of Self-compassion: Self-judgment (e.g., "I'm disapproving and judgmental about my own flaws and inadequacies."); Isolation (e.g., "When I think about my inadequacies it tends to make me feel more separate and cut off from the rest of the world."); and Over-identification (e.g., "When I'm feeling down I tend to obsess and fixate on everything that's wrong."). Similar to positive aspects, Self-judgment scale consists of five items and Isolation and Over-identification consist of four items each.
Each statement is rated on a 5-point Likert scale ranging from 1 (almost never) to 5 (almost always) (Neff, 2003b). For our study, we used the preliminary Slovenian version of the SCS. For calculating subscale scores we used the mean of subscale item responses. Before calculating a total self-compassion score, we reversed scores (i.e., 1 = 5, 2 = 4, 3 = 3, 4 = 2, 5 = 1) for subscales Self-judgment, Isolation, and Over-identification, because those three subscales measure negative aspects of Self-compassion. For a total Self-compassion score, we computed the mean of all six subscales. . The SWLS is a short 5-item instrument, which measures global cognitive judgments of satisfaction with one's life. Each statement is rated on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). The SWLS was reported having good psychometric properties, including high internal consistency and test-retest reliability (α = .87) (Diener et al., 1985). In our sample Cronbach's alpha reliability coefficient was .87.
Well-being Index (WHO-5; World Health Organization, 1998). Well-being index is a self-report questionnaire which includes five positive items, related to positive mood (relaxation), vitality (being active), and general interests (being interested in things). Items are rated on a 6-point Likert scale from 0 (not present) to 5 (constantly present). Scores are added together, with the raw score ranging from 0 to 25, with higher scores meaning better well-being. WHO-5 was shown to have adequate validity in screening for depression and in measuring outcomes in clinical trials (Topp, Østergaard, Søndergaard, & Bech, 2015). Cronbach's alpha reliability coefficient on our sample was .84.
Relationship Questionnaire (RQ; Bartholomew & Horowitz, 1991). RQ is a four-item questionnaire designed to measure adult attachment styles. Each style is rated on a 7-point Likert scale ranging from 1 (disagree strongly) to 7 (agree strongly). Attachment styles assessed are: Secure, Preoccupied, Fearful-Avoidant, and Dismissing-Avoidant style. Individuals with Secure attachment style have a sense of worthiness and an expectation that other people are generally accepting and responsive. Preoccupied attachment style shows in feelings of unworthiness combined with a positive evaluation of others. Such individuals are preoccupied with seeking emotional closeness. Individuals with Fearful-Avoidant style have a sense of unworthiness combined with an expectation that others will reject them. Despite the need for contact and intimacy, they are experiencing distrust and fear of rejection. Dismissing-Avoidant attachment style shows in avoidance of intimacy, self-reliance, and independency.

Psychometric Analyses
In preliminary analyses data were checked for normality. The total scale and subscales data were normally distributed as assessed by histograms, box plot, and levels of Skewness and Kurtosis. Data were also checked for missing values. Participants with more than three missing values were excluded from further analysis. As only 0.2% missing values remained, we replaced them with item means.
Confirmatory factor analyses were conducted to examine the factor structure of the SCS in the Slovenian sample. Based on previous research on the SCS, we compared the onefactor, the two-factor correlated, the six-factor correlated, the higher order, and the bi-factor models. Lisrel software version 9.2. was used to conduct the analysis. We used the Maximum Likelihood Robust (MLR) estimation to estimate parameters, as item responses on the SCS are ordinal ranging from 1 to 5. The same method was also used in recent research by Neff et al. (2017). Models were assessed with the following parameters: normed chi-square statistics (χ²/df), comparative fit index (CFI), non-normed fit index (NNFI), root-mean-square error of approximation (RMSEA) with accompanying 90% confidence interval (CI), standardized root-mean-square residual (SRMR), and the Akaike information criterion (AIC). In order to compare results with the original version of the SCS, we used the same liberal fit criteria used in recent studies by Neff et al. (2017) and Williams et al. (2014). An acceptable fit is obtained when the CFI and NNFI values are .90 or higher, the RMSEA is .10 or lower. Normed chisquare lower than or equal to 3, and SRMR of .10 or lower would also indicate an acceptable fit (Schermelleh-Engel, Moosbrugger, & Müller, 2003). We also documented the Akaike's information criterion (AIC) in order to compare different models between themselves. Models that have a lower level of AIC would most likely cross-validate in subsequent samples and show a better model fit. To compare results of the bi-factor model to the results of Neff et al. (2017), we calculated omega indexes based on estimates of the bi-factor model. The omega index indicates the percentage of variance in the total score accounted for by both the general self-compassion factor and subscale factors (Rodriguez, Reise, & Haviland, 2016). Omega hierarchical (ω h ; McDonald, 1999) indicates the percentage of variance in the total score that is attributed to a general self-compassion factor. The relative omega index indicates the percentage of reliable variance (not due to error) of the general factor accounted for in the total score. The Relative Omega is ω h divided by ω. Omega indexes were calculated by the Omega computer software (Watkins, 2013).

Results
Descriptive statistics and reliability coefficients for the long and short versions of the SCS are presented in Table 1. Cronbach α coefficients for SCS-LF subscales were between .66 and .84. Subscales Self-kindness, Selfjudgement, Isolation, Over-identification, and Mindfulness had acceptable to good internal reliability. The lowest reliability coefficient was obtained for the Over-identification subscale (α = .66). The reliability of the total score of the SCS was very good (α = .91).
We also evaluated reliability of the short version of SCS. The reliability of the SCS-SF total score was good (.81). Cronbach α coefficients for SCS-SF subscales were between .45 and .72. Majority of subscales had low and unsatisfactory reliability. The lowest reliability was obtained for the subscale Over-identification (.45). Mindfulness is the only subscale that had acceptable reliability.  Table 2 shows intercorrelations between subscales and the total score for both long and short form of the Slovenian SCS. The correlation between total scores of both versions was very high (.96). Table 2 Correlations between subscales and the total score for the long form (LF) and the short form (SF) Correlations between corresponding subscales of the long and short forms of SCS are presented on the diagonal. All correlations were significant at the .001 significance level. Table 3 shows fit indices for the SCS-LF that were tested by CFA. Indices in italic meet the suggested liberal cutoff criteria for having an acceptable fit. CFA of the long version indicated that the one factor (Model 1), the two factor (Model 2), and the higher order (Model 5) models did not fit the data well. Six-factor (Model 3), bi-factor (Model 6), and their re-specified versions (Models 4 and 7) showed acceptable to good fit. These models also showed the best fit compared to other models. The AIC values suggested that the respecified six-factor model would likely cross-validate the best, followed by the re-specified bi-factor model, the bi-factor model, the six-factor correlated model, the higher order model, the two factor model, and finally the one factor model. Note. SB χ² = Satorra-Bentler scaled chi-square; CFI = comparative fit index; NNFI -nonnormed fit index; RMSEA -root-mean-square error of approximation; CI = confidence interval; SRMR -standardised root mean square residual; AIC = Akaike's information criterion; Values in italic meet the suggested cutoff criteria for having acceptable fit.
The six-factor correlated model (Model 3) showed an overall satisfactory fit to the data. Indicators χ 2 /df, RMSEA, and SRMR suggested a good fit of the model to data. However, values of CFI and NNFI were below the thresholds typically used to represent an acceptable model fit. Based on modification indices we re-specified the model. The re-specified six-factor model (Model 4) includes two pairs of correlated residuals (item 5 with item 12 and item 7 with item 10). Re-specification was done with justification that these pairs of items share similar content and belong to the same subscale (Self-kindness and Common Humanity). Re-specified six-factor model had an adequate fit to the data on all indices. AIC values also indicated that the re-specified model showed a better fit than the original model and the best fit compared to all other models. We also explored the possibility of omitting some of the items, which correspond to correlated residuals. Omitting these items from SCS did not improve model fit substantially and resulted in lower subscale reliability. Table 4 shows standardised factor loadings of the six-factor correlated model of individual SCS items on their intended subscales. All loadings were statistically significant (p < .001) and ranged in magnitude from .41 to .89. The majority of items had adequate factor loadings (> .50).
Besides the six-factor model, we found that the bi-factor model (Model 6) is also acceptable. The bi-factor model showed an overall satisfactory fit, with the exception of the CFI, which was marginal and the NNFI, which was below the threshold of an acceptable fit. With the help of modification indices, we re-specified the model. The re-specified bi-factor model (model 7) includes two pairs of correlated residuals (item 1 with item 2 and item 23 with item 26). Re-specification was done with justification that these pairs of items shared similar content. Item 23 and 26 belong to the subscale self-kindness and item 1 and 2 both describe the emotional states connected to self-judgement and focusing on negative events. We also explored the possibility of omitting some of the mentioned items. This, however, did not result in a better model fit. Table 4 also shows standardized factor loadings of individual SCS items on the general self-compassion factor in the bi-factor model. Factor loadings ranged from .19 to .69 and were all statistically significant (p < .001). The majority of items had factor loadings > .40. The lowest were factor loadings on the Common Humanity Scale, where items 7 and 10 had factor loadings < .30 on the general self-compassion factor. We also calculated omega indexes based on estimates of the bi-factor model (ω = .94, ωh = .85, relative ω = .90). The Omega index indicated that the large amount of total variance in the total score could be attributed to both the general self-compassion factor and to six subscale factors. The ω index of .85 indicated that a large majority of variance in the total score can be accounted for by the self-compassion factor. The relative omega index indicated that the general self-compassion factor accounts for 90% of the reliable variance (not due to error) in the total score. Note. SF = factor loadings of items on their intended subscale in the six-factor correlated model; GF = factor loadings of items on the general self-compassion factor as found in the bi-factor model.
We also performed a confirmatory factor analysis of the short version of the SCS. Results showed that only the six-factor correlated model (Model 3) had a good fit to the data with all of the estimators above the thresholds of acceptable fit (see Table 3). Other models did not show a good fit to the data. We also examined the bi-factor model, but it didn't converge.
We tested the convergent validity of the SCS by computing the correlation between the SCS and measures of attachment styles, well-being, and satisfaction with life (Table 5). Overall self-compassion and positive self-compassion scales are positively correlated with a secure attachment style, emotional well-being, and satisfaction with life. On the other hand, lower self-compassion is related to preoccupied and fearful-avoidant attachment styles, lower emotional well-being, and lower satisfaction with life. There were no significant correlations between the dismissive-avoidant attachment style and the SCS. Correlations between the total score of the short version of SCS and examined constructs were similar to long version, although somehow higher (except for correlations with dismissiveavoidant style, where correlations were equal). Table 5 Correlations between subscales and the total score of attachment styles,and satisfaction with life (N = 255) Dimension

Discussion
In this study, we examined the psychometric properties of the Slovenian adaptation of SCS-LF/SF. In line with previous research (Neff et al., 2017;Tóth-Király et al., 2016;Williams et al., 2014), we tested different models to examine the internal structure of the SCS-LF. Specifically, we compared a one-factor, a two-factor, a six-factor correlated model, a higher order model, and a bi-factor model. We used liberal fit criteria to compare results with the recent work of Neff and colleagues (2017). Confirmatory factor analysis showed that both the six-factor and the bi-factor model have acceptable to good fit and that is better than the fit of other tested models. Neff and colleagues (2017) and Cleare and colleagues (2017) similarly found the fit of the six-factor and the bi-factor model to be better than the fit of other tested models. Similar as in other studies, the one-factor and the higher order models showed a relatively poor fit to the data (Neff et al., 2016;Williams et al., 2014). The two-factor model, which postulates that positive and negative items of self-compassion form two separate factors, also had a poor fit, similar to the study of Neff and colleagues (2017).
The six-factor correlated model showed an overall acceptable fit to the data. All indicators except CFI and NNFI indicated good fit. Similar results were also obtained in some other studies, where CFI and NNFI were below .90 (Williams et al., 2014). Based on the modification indices, we tested a respecified model, which showed a good fit on all indicators. The re-specified model includes two correlated residuals, which were also found in a Portuguese study on a nonclinical sample (Castilho et al., 2015). These residuals correspond to items 5 and 12 (kindness), and to items 7 and 10 (common humanity). These pairs of items have similar content and belong to the same factor. It is possible that these results are the consequence of adaptation of the SCS to Slovenian context and that this is somehow shared with Portuguese culture as well. The results of confirmatory factor analysis support the assumption that selfcompassion can be conceptualized as consisting of six distinct and correlated factors on the Slovenian sample.
Research of Williams and colleagues (2014) suggested that the SCS may be adequate for measuring six factors and not the total score of selfcompassion. Their conclusion was based on the finding, similar to our study, that a hierarchical model was not acceptable. Neff and colleagues (2016) proposed that the bi-factor model is more suitable than hierarchical for conceptualizing a general self-compassion factor in addition to six separate subscale factors.
In our study, we found that the bi-factor model showed an overall satisfactory fit. All indicators showed a good fit with the exception of CFI, which was marginal and NNFI, which was below the threshold of an acceptable fit. The re-specified bi-factor model, with two pairs of correlated residuals, showed a good fit with the exception of NNFI. Our results are similar to Neff and colleagues (2017) who obtained a good fit of the bi-factor model according to CFI, RMSEA, and SRMR indexes in student, community, and meditator samples. Similar to our results, they found the TLI index (NNFI) to show a marginal fit in the student and community samples, and an inadequate fit in the clinical sample. Omega values in bi-factor model indicated that large amount of variance in total score can be attributed to both the general self-compassion factor and six subscale factors. In our study, we have found approximately the same values of Omega indexes as in the study by Neff and colleagues (2017). These findings show support for using the total SCS score as well as six subscale scores.
Loadings on the general factor were generally satisfactory, however, items 7 and 10 of the scale Common Humanity had lower loadings. These items also correspond to correlated residuals found in the six-factor re-specified model. Omitting one or both items resulted in unsatisfactory reliability of the subscale, so we decided to keep those items in the SCS.
The reliability of the Slovenian version of SCS-LF was generally satisfactory, except for the subscale Over-identification. While the reliability of some subscales was somehow lower than in some other studies (Neff, 2003b), the total score reliability was very good. This indicates that the total score of Self-compassion can be used with confidence.
We also explored the convergent validity of the SCS. The Slovenian version of the SCS correlates in a predictable way with attachment styles, emotional well-being, and satisfaction with life. Higher self-compassion is positively correlated with a secure attachment style, and negatively with preoccupied and fearful-avoidant attachment. A positive correlation with secure attachment was expected, as self-compassion is defined as kindness toward self, as well as a cognitive understanding that personal shortcomings are shared with other people. These aspects of self-compassion are probably related to a positive model of self and others, which is a property of the secure attachment style. On the other hand, preoccupied and anxious-avoidant attachment styles imply a negative working model of self, and reaction with hyperactivation in attachment situations. Both these characteristics are congruent with the description of lower self-compassion, which manifests in self-judgment, isolation, and overidentification with painful emotions and thoughts. These findings are consistent with Neff and McGehee (2010), who also found positive correlations between the secure attachment style and self-compassion, and negative correlations of selfcompassion with preoccupied and fearful-avoidant attachment styles. Wei and colleagues (2011) also found a negative association between attachment anxiety and self-compassion. In our research, we didn't find significant correlations between self-compassion and the dismissive-avoidant attachment style, which is also congruent with findings that attachment avoidance doesn't significantly correlate with self-compassion (Neff & McGehee, 2010). Wei and colleagues (2011) found mixed results regarding correlations between self-compassion and avoidant attachment, and concluded that this association needs further research.
We also found that self-compassion is associated with greater satisfaction with life and emotional well-being. Our results are congruent with previous findings, which also found positive correlations with life satisfaction (Neff, 2003b(Neff, , 2015Neff, Pisitsungkagarn, & Hsieh, 2015;Wei et al., 2011), and positive affect .
We also explored the psychometric properties of the short version of the SCS. Confirmatory factor analysis of SCS-SF confirmed a satisfactory fit of six-factor model to data, and an unsatisfactory fit of other models. Total score reliability of SCS-SF was good, however similar to the study of Raes and colleagues (2011), subscales showed lower reliability. Even though we didn't find support for the higher order or the bi-factor model, we can still argue for the usability of the total score based on the good reliability and the near perfect correlation with the long form. We suggest that a shorter version of the SCS may be a good and economical alternative to the long version, when the main aim is to measure only the total self-compassion score. However, when the aim is to explore specific components of self-compassion, we, similarly as Raes and colleagues (2011), recommend the use of the longer version.

Limitation of the Present Study
The results of the study should be interpreted with consideration for some limitations. The sample is mainly comprised of females (72%). The majority of participants also had higher education. In future studies we recommend a more heterogeneous sample regarding sociodemographic characteristics. It would also be useful to explore the SCS scale on different samples, such as a clinical sample, adolescents, and the elderly.

Conclusion
We can conclude that both the long and short forms of the Slovenian version of the SCS have generally acceptable to good psychometric properties. Our results support the use of both the total score and subscale scores of the