Adapting and Examining the Factor Structure of the Self-Compassion Scale in Indonesian Version

The study aimed to adapt and analyze the factor structure of the scale of selfcompassion in the Indonesian language. The initial steps included back-to-back translation, focus group discussion and expert judgements. From this procedure, thirteen additional items were added to the original self-compassion scale. The thirty-nine final items were tested to 483 samples from three different groups (undergraduate students, young and adult, mid and late adult). Findings support self-compassion as total score and six sub scale score based on ESEM bi factor analysis. However based on the overall results, for Indonesia uses, the authors suggested the structure data of self-compassion scale is a hierarchical two-factors model with the final items of this scale are 35 items, where the two factors are positive aspect and negative aspect. The positive dimension had 0.901 of Cronbach’s Alpha reliability and the negative dimension had 0.913 of Cronbach’s Alpha reliability. The positive one consists of self-kindness, common humanity, and mindfulness. Meanwhile, the negative one consists of self-judgement, isolation, and overidentification.

The definition 1 of self-compassion is related to the more general definition of "compassion. "Compassion involves being open to and moved by the suffering of others, so that one desires to ease their suffering". It also involves recognizing that all humans are imperfect; make mistakes; show patience, kindness and nonjudgmental understanding to others. This is derived from Buddhist traditions as an alternative construct of a healthy attitude toward oneself (Neff, 2003a).
Neff developed self-compassion scale to measure three main components of selfcompassion on separate subscales (selfkindness versus self-judgment, common humanity versus isolation, and mindfulness versus over-identification), with the intention of summing the subscale scores to create a total score that would represent a participant's overall level of selfcompassion (Neff, 2003b). Neff and Germer (2013) updated self-compassion scale elements such as self-judgment which involves being harsh and extremely selfcritical; whereas self-kindness refers to being caring, understanding and accepting of oneself. Common humanity refers to the recognition that failure, disappointment, and hardship are shared human experiences; which foster connectedness to others rather than leaving one feeling isolated when faced with suffering. Mindfulness represents the acceptance of the present experience and involves taking an objective stance on one's experience in order to gain perspective and to avoid over-identification with negative thoughts and emotions.
Empirical research showed that selfcompassion is associated with psychological advantage and might be considered an important protective factor against the development and maintenance of several mental disorders such as depression and anxiety (Longden & Proctor, 2012). Selfcompassion has been linked to decreased thought suppression, anxiety, and depression, and enhancing emotional coping skills. Self-compassion has also been proven as a stronger predictor of healthy functioning than self-esteem (Neff & Vonk, 2009).
Self-compassion is also beneficial for predicting psychological resiliency in young adults and adolescents. Empirical evidence found that self-compassion can help adolescents dealing with suffering from negative self-views (Neff & McGehee, 2010). Previous research demonstrated the usefulness of selfcompassion in helping students to address failures in academic achievement (Neff, Hsieh, & Dejitterat, 2005). These two studies proved self-compassion can be useful not only for adults, but also for students and adolescents. Neff (2016) updated the definition of self-compassion to an interplay of positive and negative constructs, including to pay attention to suffering (in a mindful or over-identified way), different ways that individuals emotionally respond to pain and failure (with kindness or judgment), and to cognitively understand their predicament (as part of the human experience or as isolating). This arises as there is criticism regarding the validation of the scale of self-compassion against the definition of self-compassion, whether self-compassion scale is a valid measurement for self-compassion.
From previous studies there were disagreements whether SCS is a dimensional or multi-dimensional. This implies how to apply SCS to a population, whether to use a total score or six sub scales score. In a study in Germany, the more precise is to use SCS in two factors, namely SCS negative and SCS positive (Coroiu et al., 2018). The critique of the SCS is that the negative items of the Selfcompassion Scale (SCS), which represent reduced uncompassionate self-responding, are redundant with neuroticism and do not contribute to the concept of selfcompassion and should be disposed. Responding to this, Neff, Tóth-Király, and Colossimo (2018) re-analyzed the existing data using the ESEM bifactor analysis. The result was self-compassion scale can be measured by the total score and six subscale score.
In Indonesia there are 77 research items (based on Google Scholar searches with the keyword "influence of selfcompassion") and 71 research (based on a search in Google Scholar with keyword "self-compassion"). It shows considerable interest among researchers in Indonesia to research self-compassion. Most of them linked self-compassion with other variables related to psychological wellbeing such as loneliness, resiliency, emotional competence and so on.
There are other models and measures of self-compassion, and a lack of consensus in the field on how to define or measure compassion for self or others (Gilbert et al., 2017;Strauss et al., 2016). In Indonesia, since 2016, most researchers use a scale of self-compassion based of Neff's construct to measure compassion for self. Item number variations ranging from 12 to 26 items (Sugianto, Suwartono, & Sutanto, 2020;Khumas & Lukman, 2019;Ariyani & Hadiani, 2018, Hidayati, 2016. We found one researcher using different measuring instruments in measuring compassion; namely compassionate love scale known as the Santa Clara Brief Compassion Scale (SCBCS) developed by Hwang, Plante, and Lackey (2008). The reason for this use was because it is more appropriate as measurement for compassion in educational and religious institutions. Another thing because compassion is measured by measuring compassion to others (Arli & Anandya, 2018).
Research on self-compassion in Indonesia varies from the adult population (Febrinabilah & Listiyandini, 2016;Hidayati, 2016;Aldyafigama, baihaqi, & Pujasari, 2018), the elderly (Kistyanti & Retnowati, 2017;Karmiyati & Wahyuningsih, 2019), and adolescents (Hidayati, 2016;Ramadhani & Nurdibyanandaru, 2014;Hasanah & Hidayati, 2017;Septiyani & Novitasari, 2017). There is a variation of use of the scale self-compassion in these eight studies. The number of self-compassion items utilized used among these eight studies varied from 18 to 27 items. The lowest reliability number on the selfcompassion scale instruments used among these eight studies was 0.71 and the highest was 0.917. These eight studies generally examined the role of selfcompassion in enhancing the ability of individual's psychological wellbeing. These eight studies showed that selfcompassion in Indonesia is used in a variety of samples ranging from adolescents to elderly people.
In Indonesia there are few studies which aimed to adapt and examine the factor structure of the self-compassion scale. Only one research conducting validity and reliability analysis study on self-compassion scale in Indonesian version. The researcher named the Indonesian version of self-compassion scale with the term SWD (Skala Welas Diri) (Sugianto et al., 2020). This research sought to make cross-cultural adaptation to self-compassion scale to be used in Indonesia.
Cross-cultural adaptation is the process of translating an item by adjusting it to the cultural context in which the measuring instrument will be applied. This process includes the translation process and, if necessary, replacing the item or scaling to make it relevant and valid in the new culture (Beaton, Bombardier, Guillemin, & Ferraz, 2000). The process of adapting and examining the factor structure of self-compassion scale generally uses the CFA method with general single factor i.e., self-compassion (Tóth-Király, Bőthe & Orosz, 2017). In the case of measurements in certain populations such as measuring self-compassion in medical students, a measurement is performed on one factor and two factors. The CFA results in this study indicated that the two factor model (formed by three positive and three negative components) indicated slightly better fit than the single factor model and for the single general self-compassion factor had a borderline acceptable fit (Babenko & Quo, 2019). From the results of the adaptation there are several models that appear, but the most commonly discussed is whether the self-compassion has a factor that applies in general or has two factors. According to Tóth-Király et al. (2017) we need more sophisticated approach of analyzing the self-compassion scale to answer it. The ESEM Model is considered to be better on examining the factor structure of selfcompassion scale. In the context of adapting the self-compassion scale to a new culture, the ESEM model is considered to be more suitable for use because it can simultaneously measure factor structure by allowing cross-loading and simultaneously seeing confirmation factor analysis.
The research in this study used the CFA and ESEM bifactor models to obtain the most significant SCS adaptation in Indonesia. It is also used to see if there is a significant model difference between the population in Indonesia and the global model.

Procedure
This study was initiated by the first author and organized by all authors, who wanted to examine the factor structure in Indonesia version. Research began by reviewing the literature on self-compassion and its measurements. After obtaining theoretical reviews of selfcompassion and measurement of SCS, researchers made scale adaptations using forward and backward translation techniques (International Test Commission, 2017). The research team conducted a direct translation of the SCS by using a group discussion of six people who understand English and psychological science. Then, researchers conducted reviews and corrections to translation results. After obtaining the most contextual translation, the Indonesian translation was then sent to four experts for expert assessment.
Existing data were then processed statistically following the directions of the inventor of the scale according to the instructions written in the journal entitled "Examining the Factor Structure of the Self-Compassion Scale in 20 Diverse Samples: Support for Use of a Total Score and Six Subscale Scores" by the second author. In this part, the authors examined the factor structure of the data under two models, namely ESEM and CFA. Both in ESEM and CFA, we tested six models, namely the 1 factor CFA & ESEM (  Figure 6). In terms of model fit, the authors used three fit indices to asses model fit, namely the Root Mean Square Error of Approximation (RMSEA) (Steiger & Lind, 1980), Tucker and Lewis Index (TLI) (Tucker & Lewis, 1973) and Comparative Fit Index (CFI) (Bentler, 1990). According to Joreskog, Olsson, and Wallentin (2006) the value of CFI ad TLI will lie between 0 and 1. However, since these indices are much better for comparing the model fit to a baseline model, the good model will has a CFI and TLI value close to 1 (Asparouhov & Muthen, 2009). Or at least minimum 0.95 indicating reasonable model fit (Thompson, 2004). For the RMSEA, values of roughly 0.06 or less are generally taken to indicate reasonable mode fit (Steiger & Lind, 1980;Thompson, 2004). The authors used these three model fit statistics to evaluate whether a model is reasonably fit or not. We used Mplus 8.4 (Muthen & Muthen, 2017) to analyze the data according to our hypothesis's models.

Instruments
Appropriate approval from Institutional Review Board (IRB) at Mercu Buana University was received before collecting these data. In data processing the first author did correspondence with Kristin Neff via email to get permission adapting and validating self-compassion in Indonesian language. Email replies were received by the first author on May 9, 2020 with suggestion to use the procedure contained in the journal Neff et al. (2019) entitled "Examining the Factor Structure of the Self-Compassion Scale in 20 Diverse Samples: Support for Use of a Total Score and Six subscale Scores.
All the items were constructed in favorable wordings according to each dimension. Note that all the items in the Mindfulness subscales, Common Humanity and Self-Kindness are positively correlated with self-compassion. Meanwhile, all the items under the Self-Judgement, Isolation and Overidentification subscales are negatively correlated with selfcompassion. From Neff's study in 2003(Neff, 2003b the internal reliability of the SCS scale was 0.92 (Cronbach Alpha = 0.92) with the range of internal consistency reliability for each subscale from 0.75 to 0.81. Furthermore, the test-retest reliability over three weeks interval was satisfying with a Cronbach Alpha 0.93 for the SCS scale, and for the subscales were ranging from 0.75 to 0.81. Moreover, also from Neff's study in 2019 (Neff et al., 2019) they found that the total scale has 0.96 of omega (ω) reliability estimate and the reliability of each subscale were ranging from 0.67 to 0.84.

Participants
The number of participants was 483 from 3 groups, 124 undergraduate students (18 -24 years old), 337 emerging adults (18 -41 years old), and 22 mid and late adult age (42 -60 years). In total, the final sample is 483 respondents included 174 males and 309 females (M = 3.374, SD = 0.671). All participants were recruited by sharing online scale links in Google Forms. To investigate the factor structure of a test instrument, the recommended minimum observations are 300 samples (International Test Commission, 2017;Wolf, Harrington, Clark, & Miller, 2013), therefore the number of samples on this research was sufficient to see the factor structure of a self-compassion scale.

Results
From the expert examination, there are some items that need to be improved. Items containing the word "kindness" need to get an adjustment with the proper Indonesian language because the word "kasih" can be ambiguous with "kasih" for Christians in Indonesia. Items that contain the word "tough" cannot be replaced "keras" because it is less understandable to the community with lower education level (in the context of Indonesia, it is not obliged to study for 9 years; elementary school and junior high school). Items with the word "judgment" should not be translated with the word "menghakimi/penghakiman" but replaced with the word "menyalahkan diri". Items with sentences containing "isolation" and "suffering" need to be explained better because the meanings of suffering and isolation are not the same for every society in Indonesia. From the results of the expert examination, we added 13 items to the original self-compassion scale and 39 final items that were ready to be tested as seen at appendix A.

Structural analysis
We first tested the fit of the one-factor model of CFA and ESEM. In this model, there was only one latent variable (selfcompassionate) and 39 items. Results from Table 1, showed that both models did not fit adequately. The one-factor CFA and ESEM had CFI with 0.566 and 0.605; TLI = 0.543 and 0.583; RMSEA 0.139 and 0.133, respectively. Table 1 presents the model results of one-factor CFA and ESEM as follows. From Table 1, although the value of factor loading for each item was slightly different, the direction (positive or negative dimensions) of the loading was consistent over the two models. These factor loadings were reported in standardized coefficients. The two models had similar results both in statistical fit indices and factor loadings as well.
The next model we tested was twofactor CFA and two-factor ESEM. The two factors consisted of positive and negative aspects. The positive aspect included common humanity, mindfulness and selfkindness. The negative aspect consisted of isolation, over-identification, and selfjudgement. The results are as follow.  We evaluated the two factor models based on the information in Table 2 and 3, both CFA and ESEM had better fit compared to the one-factor model. The CFI and TLI index of two-factor CFA were 0.835 and 0.826 respectively, and the twofactor ESEM had CFI and TLI 0.829 and 0.809, respectively. The RMSEA index of two-factor CFA and ESEM were 0.086 and 0.09 respectively. Overall, the two-factor CFA had slightly better fit indices than two-factor ESEM. However, these two models were not fit according to the fit criteria from Bentler (1990), Tucker and Lewis (1973), Steiger and Lind (1980). From this model, the authors considered that self-compassion consists of two aspects; namely positive and negative aspects. This assumption was supported by the data, where the CFA model had slightly better fit indices than the ESEM model. For the next model, we tested the six correlated factors, namely common humanity, self-judgement, self-kindness, over-identification, mindfulness and, isolation. Each factor has their items, respectively. Moreover, all the factors were confirmed to correlate with each other. The results are as follow.   value on the left side is the correlation of six-factor CFA, and the right side is six-factor ESEM As can be seen in Table 4 these two models have small differences in fit indices. The correlated factor model of CFA has slightly increased in CFI and TLI with 0.915 and 0.908, respectively. Meanwhile, the correlated factor model of ESEM had slightly larger CFI and TLI with 0.95 and 0.929, respectively. The RMSEA value of the second model was 0.055 compared to the first model with 0.133. The authors calculated chi-square differences between CFA 3 rd and 5 th model and found the differences was significant. It can be concluded that the two CFA models had improved and statistically significant, in which the latest model had better fit indices (Bentler, 1990;Steiger & Lind, 1980;Tucker & Lewis, 1973). For the ESEM model, according to the Table 2 and 4, the 4 th and 6 th model had completely different results of fit indices and statistically significant. The latest model had better fit indices than the 4 th model. In the 6 th model, majority of items from all intended factors had sufficient value of factor loadings. Therefore, all the items were loaded perfectly to each factor.
From Table 6, all six factors were correlated moderately. The lowest correlation was -0.14 from the correlation between self-judgement with self-kindness, and common humanity with selfjudgement. The highest correlation was 0.705 between over-identification with isolation. From this standpoint, we can see that there are two pattern correlations between the six factors. The factors like common humanity, mindfulness, and selfkindness were correlated positively. Meanwhile, among self-judgement, isolation and over-identification had a positive correlation as well.
The next model we examined was the bifactor model with 1-g and six factors model. We run this model with CFA and ESEM analysis. The 1-g factor was a general factor that considered as a selfcompassionate, and six factors models were the same as the previous model. The results are as follow. According to Table 7, the 7 th model did not fit with the CFI and TLI 0.817 and 0.796, respectively. These indices were too far from the minimum index (0.90) (Thompson, 2004). Even the RMSEA value showed that the model was not fit (RMSEA = 0.093) (Steiger & Lind, 1980). However, the 8 th model had better fit with CFI and TLI 0.963 and 0.944, respectively. The RMSEA value was relatively small with 0.049. It indicated the ESEM analysis for 1-g factor and six-factors was fitted to the data. However, in Table 8, there were two items that not significant in factor selfkindness, that is item 1 and item 2. Moreover, item 1 has negative loading on factor self-kindness.
The next model we tested was two general factors that consisted of positive and negative aspects and six specific factors (common humanity, selfjudgement, self-kindness, mindfulness, isolation, and over-identification). We tested this model under CFA and ESEM analysis. The results of this model as follow.  In Table 9, the model number 9 with two-g's and six-factors CFA had no satisfying fit indices with CFI and TLI 0.843 and 0.826, respectively. The RMSEA index was 0.086 and it was slightly smaller compared to the previous model. The 10 th model had much better fit indices with CFI and TLI 0.963 and 0.944, respectively. The RMSEA value was 0.049 and it indicates that the ESEM factor analysis had a model fit better than the CFA analysis (Steiger & Lind, 1980). According to Table 10, all the factor loadings in model Two-BiFactor CFA, from both positive and negative aspects, were showing positive correlation with each factor. However, when the similar data was analyzed under Two-BiFactor ESEM, we obtained the correlation between some items of common humanity and mindfulness with the positive aspect were negative. It contradicted with the CFA analysis. Even though, the ESEM analysis had better fit model, the factor loadings of these two aspects slightly did not align with the factor. However, in the specific factor, all the items were positively correlated with the targeted factors.

Self-Kindness
Since all the models did not show reasonable fit, the authors proposed another model based on the CFA and ESEM results from model 3 rd , 4 th , 5 th and 6 th . From these models, the authors obtained insight about structure of the data where there were two correlation factors namely negative and positive aspects and for each aspect consists of three dimensions. Since the correlations between dimensions were moderate, the author proposed a hierarchical two-factor model. The first level was three dimensions for each aspect and the second level was two aspects (in Figure 6). However, we deleted some items with factor loadings that were not significant (p > 0.05). Therefore, we only examined items with significant factor loadings. Items that we deleted were item number 3, 4, 9 of self-kindness and item 16 of common humanity. So, the authors examined these models only under CFA analysis because ESEM analysis cannot estimate a latent variable. Based on Table 11, the hierarchical two factors CFA had good fit indices with RMSEA 0.059, CFI and TLI were 0.935 and 0.93. It indicated that these models had reasonably fit well to the data (Bentler, 1990;Tucker & Lewis, 1973). Also, all the items were load highly on each factor. The lowest factor loading of item was 0.579 and the highest factor loading of item was 0.904. In the second level, as can be seen in Table 12, the factor loadings of each dimension from each factor were showing reasonably well. The self-kindness had 0.82 of factor loading, common humanity had 0.747 of factor loading and mindfulness had 0.916 of factor loading on positive aspect. Meanwhile, on the negative aspect, the self-judgement had 0.683 of factor loading, isolation had 0.923 of factor loading and over-identification had 0.90 of factor loading. These results were satisfying and reasonably fit well to the data.
Based on the overall results, the authors suggested the structure data of this study was a hierarchical two-factors model with the final 35 item, where the two factors are negative aspect and positive aspect. The positive one consists of mindfulness, self-kindness, and common humanity. Meanwhile, the negative one consists of over-identification, selfjudgement, and isolation.

Reliability
In the reliability analysis, the authors only analyzed 35 valid items (in Appendix B) with each dimension and aspect. The authors calculated the reliability analysis using Alpha Cronbach reliability. The scale is reliable if the Alpha-Cronbach index close to 1 and as minimum standard reliability is 0.7 (Nunally & Bernstein, 1994). The results are as follow. Based on Table 13, each dimension had reasonably well reliability index. The positive dimension had 0.901 of Cronbach's Alpha reliability and the negative dimension had 0.913 of Cronbach's Alpha reliability. However, for the unidimensional factor where all the items considered as a single factor or score, the reliability index decreased significantly to 0.761. These findings were consistent with the results from hierar-chical two-factors CFA and the score cannot be treated as a single score.

This
measuring instrument model supports self-compassion as a concept that can measure one G factor i.e. selfcompassion and six sub-scales. It is demonstrated from the model that is in Table 7 by using the ESEM Bi Analysis tool factor. The numbers are acceptable on the allowed threshold of RMSEA 0.049, CFI 0.963, and TLI 0.944. In this model there are only two items on self-kindness unrelated to the self-kindness construct i.e., Item No 1 and Item No 2. These result support Neff et al.'s (2019) argument that the bifactor-ESEM was the most conceptually appropriate way to interpret the SCS because it can simultaneously specify both the specific and overall relationship of items using a bifactor framework approach as well as their interaction as a system with an ESEM approach.
From the overall result we gained insight to make sharper modifications to get a better fit model. The results from our analysis proposed a hierarchical two-factor model thathat has two levels of models. The first level is three aspects of two factors (common humanity, self-kindness, and mindfulness as positive factors; isolation, self-judgment, and over identification as negative factors). The second level is two factors, positive and negative. This implicates that there are two total score factors i.e., the total negative score and the total positive score.
In bifactor studies, interpretation of the score can be determined from the fittest models of the data (Rodriguez, Reise, & Haviland, 2016a). One of the important data is the coefficient alpha, from the data in this study of coefficient alpha with two-factor hierarchical models better than the bi factor models. Neff et al.
(2019) has carried out global measurements and found evidence that selfcompassion is measured better with the total score and six sub-scale score. According to us the results of this research do not necessarily change the global selfcompassion. This research enhances the use of self-compassion especially in Indonesia. From the results of the construct and data model that we acquired; we suggest modifications in calculating the total score of self-compassion. According to Neff (2003b) to calculate the total score is by doing reverse coding on the negative aspects (self-judgment, over identification, and isolation), the results are then added to the positive aspects. The result of the summation aspect becomes the total score of self-compassion. We advise that for use in Indonesia, to get the total score of the overall self-compassion is by making a deduction between the total positive score with the total negative score.
The result of this summation will then show whether the total score of the participant's self-compassion is in the total positive score or the total negative score. The total positive score indicates that participants have good self-compassion, and a negative total score indicates a lack of good self-compassion.
These measuring models need to be tested in several special groups such as entrepreneurs and workers. Research in entrepreneurs and workers in Indonesia that have been done recently mostly discussed the quality of life (Syaiful & Bahar, 2016;Wardani & Anwar, 2019). Another research in Indonesia that is also widely cited is self-compassion and resilience in drug addicts. In Indonesia resiliency in drug addicts has been discussed separately (Syaiful & Dearly, 2015) or linked to self-compassion (Febrinabilah & Listiyandini, 2016).
The interesting thing about selfcompassion is that the construction is something that can be taught and studied (Neff et al., 2019;Neff & Garmer, 2013). The results of present study can inspire educators or therapies in Indonesia that teach self-compassion for sensitive viewing of negative and positive behaviors of the measurement results of self-compassion scale.

Conclusion
From the results of this study we suggest the use of Self compassion model in Indonesia using two-factor hierarchical models. Further research needs to be done by relating this model when applying selfcompassion intervention in Indonesia.
This research also opens up opportunities for researchers who want to measure self-compassion in Indonesia to see the consistency of items in measuring self-compassion. According to Rodriguez, Reise, and Haviland (2016b), the results of the analysis data factor structure need to be demonstrated in field research. It is a challenge for further research in Indonesia to determine which model is better when studying and teaching self-compassion in Indonesia.
The research also inspires researchers in Indonesia that self-compassion is something that needs to be defined as a form of behavior in Indonesia. The hierarchical two factor model can help researchers or psychologists who want to teach self-compassion in Indonesia by giving an understanding of the positive and negative factors of self-compassion.
This research has a limitation because of non-proportionate sample that is less proportional between adults and the mid and late adults. Further research can conduct research with better sample compositions.

Suggestions
This research has a limitation because of non-proportionate sample that is less proportional between adults and the mid and late adults. Further research can conduct research with better samples.

Appendix B Final Items
Final items in the Self Kindness Subscale No Final Item S-KI 1 Saya menghargai diri sendiri ketika saya mengalami penderitaan.

Final items in the Common Humanity Subscale
No Final Item C-HU 17 Saya mencoba melihat kegagalan saya sebagai hal yang lumrah dan terjadi pada setiap orang.