Translation and validation of the Swedish version of the Self-Compassion Scale for Youth

ABSTRACT

Objective

Compassion-focused interventions for young people have started to emerge to treat depressive symptoms, and reliable and valid measures of the construct “self-compassion” is needed for this age-group in Swedish. This study aims to validate the Swedish translation of the Self-Compassion Scale for Youth (SCS-Y).

Method

Self-report questionnaires were collected from students (N = 316) aged 15–20 recruited from schools in Sweden, in a cross-sectional design. Confirmatory factor analyses, internal consistency, test-retest reliability, measurement invariance and convergent and divergent validity were calculated.

Results

A model with one general bi-factor and six specific factors had the best fit and confirmed the factor structure of SCS-Y. Internal consistencies were good, except for the subscale mindfulness for boys which was questionable. Three-week test-retest reliability was good. We found measurement invariance for age and no equivalence for sex. Evidence was found for convergent and divergent validity using correlations. Sex differences were found: girls scored higher on self-judgement, feelings of isolation and showed more over-identification, than boys.

Conclusions

The Swedish version of SCS-Y (SCS-Y-SE) is a reliable and valid self-report questionnaire, with some limitations, to assess self-compassion in clinical practice and research.

Key Points

What is already known about this topic:

1. Compassion-focused interventions can potentially help young people overcome self-criticism and poor psychological health.

2. Reliable and valid self-report questionnaires for compassion focused interventions are needed in Swedish.

3. Self-Compassion Scale for Youths (SCS-Y) is a short, age-appropriate, valid and reliable instrument for adolescents in US populations.

What this topic adds:

1. We translated and adapted the SCS-Y to Swedish (SCS-Y-SE) and tested it in a school sample.

2. The factor structure of SCS-Y-SE was established by confirmatory factor analyses, three-week test-retest reliability was good, and convergent and divergent validity were good.

3. SCS-Y-SE can be considered a reliable and valid Swedish questionnaire for self-compassion in the age-range 15–20 years.

KEYWORDS:

• Self-compassion
• adolescents
• confirmatory factor analysis
• measurement invariance
• psychometrics

Introduction

Adolescence includes major developments in neurology, cognitive function, physical ability and emotional regulation (Susman & Dorn, 2009). Additionally, self-identity forms and relationship skills improve, and a stable and strong sense of identity and well-functioning relationships with peers are related to well-being in young people (Ragelienė, 2016). An absence of interpersonal connectedness (Caine et al., 2018) and self-compassion (e.g., Bluth & Blanton, 2015; Marsh et al., 2018; Zeller et al., 2015) is related to psychological problems and increased risk for suicide (Zeller et al., 2015).

Self-compassion is drawn primarily from Buddhist teaching in the Insight tradition (Brach, 2003) and has been defined as “compassion turned inward and refers to how we relate to ourselves in instances of perceived failure, inadequacy, or personal suffering” (Neff, 2015). Compassionate behaviours include self-compassion (being warm and understanding towards oneself), aspects of common humanity (sharing one’s own limitations and imperfections with others) and mindfulness (being aware of the presently-experienced moment without being overwhelmed). Uncompassionate behaviours include self-judgement (being hard and negative towards oneself), isolation (feeling alone when suffering) and over-identification with someone else’s suffering. These components are conceptually distinct and tap into different ways that individuals attend to, respond to, and cognitively understand suffering. They represent various dimensions of self-compassion, which interact as a system (Neff, 2015).

The result of two recent meta-analyses, one for adults (MacBeth & Gumley, 2012) and one for young people (Marsh et al., 2018), showed strong correlations between perceived self-compassion and psychological stress, anxiety and depression. High self-compassion in young people is related to increased well-being (Bluth & Blanton, 2015; Bluth et al., 2017; Henje et al., 2020).

Preliminary findings of compassion-focused therapy (CFT) (Bluth, Gaylord, et al., 2016; Boggiss et al., 2020; Bratt et al., 2020; Burke et al., 2019; Donovan et al., 2021) show that CFT can help young people overcome challenges during adolescence. However, there is a need for a reliable and valid self-report to assess the clinical outcomes of CFT, for clinical practice and in research.

Self-report questionnaires of self-compassion

One of the most studied self-compassion scales is the adult version of the Self-Compassion Scale (SCS; Neff, 2003), which is not optimal for youths due to its complex language (Neff et al., 2021). Neff et al. (2021) subsequently developed a youth version of the SCS (SCS-Y) adapted for 12–14 years olds. The 17-item SCS-Y was tested and analysed in an exploratory study, and two confirmative studies and a bifactor exploratory structural equation model (bifactor ESEM) supported the use of a general self-compassion score and six subscale scores. The validity and structure of Neff et al. (2021) SCS has been debated (Cleare et al., 2018; Ferrari et al., 2022; Muris & Petrocchi, 2017; Neff et al., 2021) and it has been questioned whether compassionate self-responding (CSR) and uncompassionate self-responding (USR) are one single dimension or two semi-independent, unipolar dimensions (Ferrari et al., 2022). This debate is not yet resolved.

CEAS-Y builds on Gilbert et al. (2017) biopsychosocial theory of compassion that investigates (in addition to self-compassion) how respondents give and receive compassion. Self-compassion was shown to be more important for psychological health (Henje et al., 2020) than the ability to give and receive compassion. At present, there is no Swedish self-report instrument that specifically targets self-compassion.

Sex differences in self-compassion

Studies about sex differences in self-compassion have shown mixed results for adults (Yarnell et al., 2015). In adolescent studies, girls scored lower overall on self-compassion than boys (Bluth & Blanton, 2015; Bluth et al., 2017; Henje et al., 2020), and interestingly, girls in late adolescence showed lower self-compassion than either boys or younger girls (Bluth & Blanton, 2015; Bluth et al., 2017). There may be several explanations for this pattern; for example, girls and boys are usually raised differently (Connell, 2009; Parpart et al., 2000). Furthermore, girls can suffer from sexual objectification and are more exposed to sexual harassment and abuse during adolescence than boys (Bradbury-Jones et al., 2019; Stahl & Dennhag, 2020). It is well known that these mechanisms can be internalised and affect self-confidence, self-worth and identity, and increase the risk for depressive symptoms e.g., (Dodd & Littleton, 2017). More self-objectification, body shame and negative eating attitudes have also been related to lower self-compassion (Liss & Erchull, 2015).

The objectives of the study

This study aimed to translate SCS-Y (Neff et al., 2021) to Swedish and validate it in a secondary school sample including age 15–20. The age span was a methodological necessity; it was the age of secondary school participants, which we intend to target with self-compassion interventions in a next phase. In this study, the term “young people” refers to 15–20 year olds (in other studies, “young people” has included ages 10–24, McDonagh et al. (2018)). In addition to confirming SCS-Y with regard to Neff’s et al. (2021) single-bifactor model, we wanted to investigate whether CSR and USR are one single dimension or two semi-independent, unipolar dimensions.

Materials and methods

Study setting and design

The study was cross-sectional and conducted in northern Sweden. Ethical approval for the research project was given by the Regional Ethical Review Board in Umeå, Sweden (number 2018/59–31)

Procedure

Permission for the translation process and for conducting the study was obtained from the author of the SCS and SCS-Y questionnaires (Neff et al., 2021). We used a back-translation method to obtain semantic and content equivalence across cultures (Hambleton, 2004). A professional translator was hired for back-translation, and a few corrections were made after that. Four adolescent girls read and answered the questionnaire and confirmed the relevance and understandability of the items, before distribution.

All upper secondary schools in the region of Umeå were invited to participate, two of which agreed. Permission for conducting the study at the schools was granted by the principals. In total, 442 students received verbal and written information about the study from their teachers and the research assistants. Written consent was obtained from those who volunteered for participation (316, 71%). The students answered all questions on a web survey, and they were rewarded with a gift certificate worth around 20 US dollars.

Participants

Participants (N = 316), of which 213 were girls (67%), constituted a convenience sample from two upper secondary schools. The age ranged from 15 to 20 years (M = 17.07, SD = 1.36). The distribution was as follows: six participants were 15 years old, 91 were 16 years old, 108 were 17 years old, 86 were 18 years old, 20 were 19 years old and 5 were 20 years old. Most participants were born in Sweden (90%) and all of them spoke Swedish fluently. A majority of the participants lived with both parents (67%), the socioeconomic status of whom were classified according to a Swedish socioeconomic classification system made by Statistics Sweden (Statistiska Centralbyrån, 1984, 2019). The distribution was as follows: 17.60% manual workers, 32.6% clerical or office workers, 32.60% professionals, higher civil servants and executives, 7.60% self-employed of various kinds and 11.10% unknown. A subset of the sample (119 girls and 36 boys, mean age 16.91 (SD = 0.84)) filled in the same questionnaires three weeks later to obtain data on test-retest reliability.

Self-report questionnaires

All self-report questionnaires were administered in Swedish.

Self-Compassion Scale for youths – Swedish version (SCS-Y-SE)

SCS-Y-SE was translated from the original SCS-Y (Neff et al., 2021) comprising17 items divided into six subscales that can be used separately or combined for a total score. The items of self-kindness, common humanity and mindfulness were rated on a 5-point Likert scale from 1 (almost never) to 5 (almost always). Self-judgement, isolation and over-identification were reversely graded. Psychometrics were satisfactory in the English version by Neff et al. (2021), who found internal consistency to be questionable to excellent (ω = .93 for the total scale; Cronbach’s alphas for the subscales: .85 for self-compassion, .71 for self-kindness, .80 for self-judgement, .72 for common humanity, .78 for isolation, .67 for mindfulness, .73 for overidentification) and test-retest reliability was satisfactory (r = .51 to .71, total score .83).

Measures to examine convergent validity

Self-Compassion Scale (SCS)

SCS (Neff, 2003) is a 26-item instrument that measures responses to difficult life circumstances and inner triggers. Compassionate responses include self-kindness, common humanity and mindfulness, and uncompassionate responses include self-judgement, isolation and over-identification. The items were rated on a 5-point Likert scale from 1 (almost never) to 5 (almost always) with higher scores indicating more self-compassion. A total mean score was calculated after reversing the items of the negative subscales (self-judgement, isolation and over-identification). The Swedish version of SCS (Norrbottens läns landsting, 2013) has been validated in a sample of students (Nicolaidis, 2018). Internationally, SCS has been validated in a large adolescent sample, ages 12–19 (Cunha et al., 2016). Internal consistency in the current sample was .88 (95% CI = [0.86, 0.90]).

WHO-Five well-being index (WHO-5)

WHO-5 (Blom et al., 2012) measures overall psychological well-being in five items: 1. being in good spirits; 2. feeling relaxed; 3. having energy; 4. waking up fresh and rested; 5. being interested in things. A 6-point Likert scale from 0 (not present) to 5 (constantly present) was used with higher scores meaning better well-being (Blom et al., 2012). The total sum score was calculated (range from 0 to 25), and the percentage value was calculated by multiplying the score by 4 and thus obtaining a scale from 0 (worst imaginable wellbeing) to 100 (best imaginable well-being). WHO-5 was found psychometrically valid in a Swedish adolescent psychiatric sample (ages 14–18) (Blom et al., 2012). Internal consistency in the current sample was .89 (95% CI = [.87, .91]).

Beck Youth Inventories (BYI)

The five subscales of BYI (Beck et al., 2001, 2017) are rated on a 4-point Likert scale from 0 (never) to 3 (always). Higher scores represent more severe symptoms. Total sum scores were calculated using only the Depression (BYI-D) and Anger (BYI-A) subscale scores. BYI was found psychometrically valid in a Swedish school sample, ages 7–18 (Beck et al., 2017). Internal consistencies were high in the present study: .92 (95% CI = [.91, .94]) for BYI-D and .92 (95% CI = [.90, .93]) for BYI-A.

Revised Child Anxiety and Depression Scale (RCADS)

RCADS long scale (Chorpita et al., 2005) measures anxiety and depression symptoms. The 47-item scale was rated on a 4-point Likert scale, from 0 (never) to 3 (always). Here, we used the total anxiety scale (37 items) and the total depression scale (10 items). Total sum scores were calculated. The Swedish version was retrieved from a homepage for RCADS (Chorpita & Spence, 1998), which was found valid in a Danish sample age 8–16 (Esbjørn et al., 2012). The internal consistencies were excellent (α = .94 and .91) in the present study.

Measures to examine divergent validity

Strength and Difficulties Questionnaire (SDQ)

SDQ (Goodman, 2001; Hagquist, 2007) measures the psychological adjustment of children and youth. Here, we used only the SDQ-impact subscale, which is considered a measure of global functioning. The five-item SDQ-impact subscale uses a 5-point Likert scale from 1 (Not at all) to 5 (All the time) to assess how much psychological problems interfere with daily life. The sum score was calculated. The internal consistency was acceptable (α = .69) (95% CI = [.61, .75]) in the present study. The SDQ has been found psychometrically valid in a Swedish adolescent sample (ages 12–18, Hagquist, 2007).

PROMIS pediatric pain interference v.2.0 (Rindestig et al., 2020)

Patient Reported Outcome Measurements Information System (PROMIS) assesses generic health for children (Cella et al., 2007). Here, the item bank for pain interference was used (www.healthmeasures.net). The 20-item measure uses a 5-point Likert scale from 1 (never) to 5 (almost always). The sum score was calculated. The internal consistency was excellent in this sample (α = .96) (95% CI = [.96, .97]). The scale pain inference was found psychometrically valid in a Swedish adolescent sample (ages 12–19, Rindestig et al., 2020).

Statistical analysis

Descriptive analyses were conducted using SPSS, Version 27.0. We analysed corrected item-total correlations (r_it ^c) and the cut-off for removing items was set to less than .30 (Field, 2018).

SCS-Y-SE’s dimensionality was tested using the Lavaan R package for structural equation modelling version 0.60–3.00 (BETA); (Rossell, 2018). Two confirmative models of SCS-Y were analysed using confirmatory factor analysis (CFA). First, a bifactor model with one global factor was estimated: items loaded on a self-compassion global (G) factor and on their corresponding six specific factors. No cross-loadings were allowed between G and specific factors (Model 1), but specific factors were allowed to correlate. Second, a two-bifactor model was calculated and no cross-loading was allowed between G and specific factors. The two global factors (compassionate and uncompassionate self-responding) were allowed to correlate with one another. Specific factors were allowed to correlate (Model 2). We tested whether CSR and UCS were single dimension or semi-independent by a Chi-square test between the models (Figure 1).

Figure 1. Graphical comparison of the two CFA.

CFA: Confirmatory Factor Analysis; Circles represent latent factors; squares represent scale items.

Robust diagonally weighted least squares (DWLS) estimator and a scaled-shifted test statistic (SS) were used due to the ordinal scale responses (Li, 2016a, 2016b). We used a polychoric correlation matrix as it is more suitable for ordinal data. The goodness-of-fit indices were Satorra – Bentler Chi-square (SB χ2), coefficient χ2/df, comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of approximation (RMSEA) and standardised root mean square residual, SRMR <0.08 (Hu & Bentler, 1999). A good model fit is obtained if RMSEA <0.05 (Browne & Cudeck, 1992), 3 or less for χ2/df (Kline, 1998) and TLI/CFI >0.95 (Hu & Bentler, 1999).

The reliability of the scales was estimated by Cronbach’s α and McDonald’s omega (ω). Internal consistency is considered to be good between .7 and .9 (Nunnally & Bernstein, 1994). Test-retest reliability (over 3 weeks) was measured by Intraclass Correlation Coefficients (ICCs). We used Koo and Li’s (Koo & Li, 2016) criteria: <.5 poor, .5–.75 moderate, .75–.9 good and >.9 excellent reliability.

Construct validity was calculated by Pearson’s correlations (r) between subscales (convergent and divergent validity).

Measurement equivalence or invariance across groups was used to assess measurement invariance across sex and age groups. As recommended by Svetina et al. (Svetina et al., 2019), we estimated the models in the following order: initial baseline models (configural invariance models), a test of invariant factor loadings (metric invariance) and invariant factor covariance (scalar invariance). A CFI change (ΔCFI) equal to or less than −0.01 is evidence of measurement invariance, even with a small sample (e.g., 150 per group) (Cheung & Rensvold, 2002; Finch et al., 2018). An RSMEA change (ΔRSMEA) of ≤0.01 can be interpreted as an excellent fit (close fit, 0.01–0.05; fair fit, 0.05–0.08; mediocre fit, 0.08–0.10; poor fit, >0.10) (Finch et al., 2018), categories from Svetina et al. (2019). A significant Satorra–Bentler test indicates that there is a difference between the models (Svetina et al., 2019). Sex differences were measured by independent t-tests (p = .05). Age differences were tested with an independent t-test.

Results

Descriptive statistics of the items

See Tables 1 and 2 for descriptive statistics of the items.

Table 1. Descriptive statistics of the total scale and subscales for total sample, boys and girls in the self-compassion scale for youth in Swedish.

	Total sample N = 316		Boys N = 103		Girls N = 213		T-test
	M	SD	M	SD	M	SD	T	p	ES d
Total scale	3.01	0.69	3.19	.67	2.92	.68	−3.32	.001**	−.40
Self-compassion total scale	3.01	0.69	3.19	0.67	2.92	0.68	−1.43	0.15	−.17
Self-Kindness	2.93	0.94	3.03	0.93	2.87	0.95	−3.87	.001***	−46
Self-Judgement (reversed)	3.11	1.03	3.43	1.08	2.96	0.97	0.47	.640	.05
Common Humanity	2.63	1.01	2.60	1.05	2.65	1.00	−3.73	.001***	−.45
Isolation (reversed)	3.39	0.96	3.67	1.03	3.25	0.90	−1.52	.065	−.18
Mindfulness	3.01	0.87	3.12	0.89	2.96	0.86	−4.77	.001***	−.57
Over-identified (reversed)	2.97	1.03	3.35	1.08	2.78	0.96	−3.32	.001**	−.40

M = mean, SD = standard deviation. Cohens d uses the pooled standard deviation. ES = Effect size d ≥ .2 for a small effect, ≥ .5 for a medium effect, ≥ .8 for a large effect.

Table 2. Descriptive statistics of the items for total sample, boys and girls in the self-compassion scale for youth in Swedish.

	Total sample N = 316				Boys N = 103				Girls N = 213
Items	M	SD	rit ^c	α-i	M	SD	rit ^c	α-i	M	SD	rit ^c	α-i
1. I try to be kind and supportive to myself when I’m having a hard time.	3.06	1.23	0.58	0.87	3.31	1.20	0.52	0.85	2.94	1.22	0.59	0.88
2. When I feel sad or down, it seems like I’m the only one who feels that way.	3.47	1.14	0.53	0.88	3.86	1.12	0.60	0.85	3.29	1.11	0.46	0.89
3. When I notice things about myself that I don’t like, I get really frustrated.	3.25	1.22	0.63	0.87	3.72	1.16	0.57	0.85	3.03	1.18	0.63	0.88
4. When I feel I’m not “good enough” in some way, I try to remind myself that other people sometimes feel this way too.	2.53	1.09	0.47	0.88	2.52	1.13	0.41	0.85	2.53	1.07	0.51	0.88
5. When I feel frustrated or disappointed, I think about it over and over again.	2.92	1.19	0.55	0.88	3.40	1.26	0.58	0.85	2.69	1.09	0.51	0.88
6. When something upsetting happens I try to see things as they are without blowing it out of proportion.	2.88	1.17	0.52	0.88	3.03	1.23	0.33	0.86	2.81	1.13	0.62	0.88
7. I get mad at myself for not being better at some things.	3.00	1.18	0.55	0.88	3.21	1.26	0.62	0.84	2.90	1.13	0.49	0.88
8. When I’m sad or unhappy, I remember that other people also feel this way at times.	2.62	1.14	0.43	0.88	2.61	1.17	0.37	0.86	2.62	1.12	0.48	0.89
9. I’m kind to myself when things go wrong and I’m feeling bad.	2.67	1.19	0.61	0.87	2.74	1.23	0.56	0.85	2.63	1.17	0.65	0.88
10. When I feel bad or upset, I tend to feel most other people are probably happier than I am.	3.46	1.15	0.55	0.88	3.74	1.24	0.55	0.85	3.32	1.08	0.52	0.88
11. When something difficult happens, I try to see things clearly without exaggerations.	3.07	1.08	0.47	0.88	3.11	1.08	0.33	0.86	3.05	1.08	0.55	0.88
12. I’m really hard on myself when I do something wrong.	3.07	1.24	0.57	0.88	3.42	1.26	0.59	0.85	2.90	1.21	0.53	0.88
13. When things aren’t going well, I keep in mind that life is sometimes hard for everyone.	2.71	1.16	0.50	0.88	2.65	1.22	0.40	0.86	2.75	1.13	0.59	0.88
14. When I’m feeling bad or upset, I can’t think of anything else at the time.	3.03	1.12	0.52	0.88	3.32	1.14	0.43	0.85	2.88	1.08	0.54	0.88
15. I try to be understanding and patient with myself even when I mess up.	2.98	1.11	0.51	0.88	3.05	1.18	0.47	0.85	2.95	1.08	0.54	0.88
16. When I’m really struggling, I tend to feel like other people are probably having an easier time of it.	3.24	1.15	0.47	0.88	3.54	1.14	0.49	0.85	3.10	1.14	0.44	0.89
17. When something upsets me, I try to notice my feelings and not get carried away by them.	3.02	1.08	0.36	0.88	3.19	1.24	0.25	0.86	2.94	0.98	0.40	0.89

M = mean, SD = standard deviation, r_it ^c = corrected item-total correlation, α-i = ordinal alpha if the item is removed.

Correlations between the SCS-Y-SE subscales

As expected, we found high correlations between the subscales. The highest correlations (medium to large) were within the compassionate subscales (self-kindness, common humanity and mindfulness) and within the uncompassionate subscales (self-judgement, isolation and overidentified) (Table 3).

Table 3. Intercorrelations between SCS-Y-SE subscales using Pearson’s correlation test.

Subscale	Self-judgement	Common humanity	Isolation	Mindfulness	Overidentified
Self-kindness	.44**	.57**	.36**	.67**	.34**
Self-judgement	1.000	.20**	.66**	.26**	.71**
Common humanity		1.000	.16**	.46**	.12*
Isolation			1.000	.26**	.57**
Mindfulness				1.000	.25**

*p < 0.05, **p < 0.01.

Confirmatory factor analyses

Model 1. A bifactor model with one global factor was estimated with CFA, and there was an excellent fit of the model to the data χ² (87) = 158.72, χ²/df = 1.82, CFI = .99, TLI = .98, RMSEA = .05, (90% CI .04, .06) and SRMR = .03. See Figure 1 for standardised factor loadings for model 1’s global factor and specific factors.

Model 2. A two-bifactor model with two global factors was estimated, and it had an excellent model fit to the data (χ² (136) = 153.51, χ² /df = 1.78, CFI = .99, TLI = .98, RMSEA = .05 (90% CI .04, .06) and SRMR = .03). A comparative Chi-square test between models 1 and 2 showed no significant difference χ² diff (86, N = 316) = 3.76, p = .052. Standardized factor loadings are shown in Figure 1.

Reliability

The corrected item-total correlations (r_it^c) were higher than 0.30 in the total sample, and in the subsamples of girls and boys, except for one item (17 in the subscale “over-identification” was lower for boys (r_it^c = .25)). We still decided to keep all items for holding the scale theoretically intact.

Internal consistency for the total sample was good both for the total scale (α = .88, ω = .87) and the subscales self-kindness, self-judgement, common humanity and isolation (α = .69–.87, ω .73–.86). The subscale mindfulness had lower internal consistency (α = .69, ω = .70). Separate analyses were made for girls (α = .89, ω = .88 for the total scale, subscales α = .70–.86, ω = .73–.86) and boys (α = .86, ω = .83 for the total scale, subscales α = .62–.88, ω = .63–.88). They were all considered at an acceptable level, except for the boys’ result for the subscale mindfulness (α .62, ω .63) which was questionable (Table 4).

Table 4. Internal consistency of the SCS-Y translated to Swedish.

	Boys (n = 103)		Girls (n = 213)		Total
	α	ω	α	ω	α	ω
Total scale	.86	.83	.89	.88	.88	.87
Subscales
Self-kindness	.68	.72	.72	.73	.70	.72
Self-judgement	.83	.84	.77	.78	.80	.80
Common humanity	.88	.88	.86	.86	.87	.87
Isolation	.82	.82	.75	.75	.78	.79
Mindfulness	.62	.63	.72	.75	.69	.70
Over-identification	.75	#	.70	#	.73	#

*= p < .05, **p < .0.01, ***p < .001.M = mean, SD = standard deviation, α = Cronbach alpha. ω = McDonald’s omega. # = Omega cannot be estimated because the number of items is less than 3.

Three-week test-retest reliability, measured by ICC, was good for the total score of the scale (.86; 95% CI .81, .89; F = 12.81; p ≤ .001). The ICCs were moderate for the subscales: self-kindness (.74; 95% CI .66, .80; F = 6.64; p ≤ .001), self-judgement (.68; 95% CI .58, .86; F = 5.25; p ≤ .001), common humanity (.73; 95% CI .65, .79; F = 6.50; p ≤ .001), isolation (.70; 95% CI .60, .77; F = 5.56; p ≤ .001), mindfulness (.74; 95% CI .65, .79; F = 6.55; p ≤ .001) and overidentified (.66; 95% CI .56, .74; F = 4.93; p ≤ .001). The test’s stability over time was moderate to good.

Convergent and divergent validity

Convergent validity between SCS-Y-SE and self-compassion (SCS), overall well-being (WHO-5), anxiety (RCADS Anx), depression (BYI-D) and anger (BYI-A), were investigated. SCS-Y-SE showed small to very high positive correlations to self-compassion (r = .21 to .78, p < .010) and well-being (r = .18 to .50, p < .010), and small to high inverse correlations to anxiety (r = −.16 to −.61, p < .010), depression (r = −.18 to −.61, p < .010) and anger (r = −.17 to −.49, p < .010), supporting the convergent validity of the scales.

The analysis of divergent validity, between SCS-Y-SE scales, and the scales measuring SDQ-functioning composite (r = −.14, p = .060 to −.29, p < .010) and PROMIS Pain interference (r = .00, p = .99 to −.21. p < .010), showed no to small associations, supporting the divergent validity of the scales (descriptive statistics of the validity scales and the Pearson correlations appear in Table 5).

Table 5. Descriptive statistics of validity measures and Pearson’s correlations between scales.

Total sample N = 316	SCS-Y-SE total scale and subscales
Scales	M	SD	TOT	SK	SJ	CH	I	M	O
Self-Compassion Scale			.72**	.43**	.78**	.21**	.74**	.25**	.68**
WHO-5 Well-being Index total scale	55.67	21.54	.48**	.32**	.50**	.18**	.38**	.25**	.45**
Beck Youth Inventories Depression subscale	17.36	10.31	−.59**	−.40**	−.61**	−.18**	−.55**	−.28**	−.53**
Beck Youth Inventories Anger subscale	10.74	8.64	−.49**	−.34**	−.49**	−.17**	−.38**	−.32**	−.40**
The Revised Child Anxiety and Depression Scale – total Anxiety scale	29.46	17.78	−.56**	−.36**	−.61**	−.16**	−.51**	−.26**	−.51**
Strength and Difficulties Questionnaire – Functioning Composite subscale	1.28	1.59	−29**	−.29**	−.23**	−.17*	−.15*	−.24**	−.14
PROMIS Pediatric Pain Interference v.2.0	29.52	13.59	−.21**	−.21**	−.19**	.00	−.18**	−.20**	−.13*

*p < 0.05, **p < 0.01. M = mean, SD = standard deviation. TOT = Total scale SCS-Y-SE, SK = Self-kindness, SJ = Self-Judgement, CH = Common Humanity, I = Isolation, M = Mindfulness, O = Overidentified.

Measurement invariance across groups

Measurement invariance for sex

The configural invariance model demonstrated a good fit to the data indicating that the factor structure of SCS-Y-SE was equivalent for boys and girls. The metric invariance model showed an acceptable fit. Two goodness-of-fit tests $\mathrm{\Delta }$CFI and the Satorra–Bentlers test indicated inequality for Configural vs. Metric and Metric vs. Scalar invariance model. The scalar invariance model showed a good fit indicating that there were no observable sex differences between the intercepts in the models (Table 6). We concluded there was an ambiguous measurement invariance across groups divided by sex.

Table 6. Measurement invariance goodness-of-fit for the 6-factor model of SCS-Y-SE presented by sex and age-group.

Invariance	$\mathit{\chi }$^2(df)	CFI	TLI	RSMEA (90% CI)	$\mathrm{\Delta }\mathit{\chi }$	$\mathrm{\Delta }\mathit{df}$	$\mathrm{\Delta }$CFI	$\mathrm{\Delta }$RSMEA
Sex
Configural	427.37** (200)	0.96	0.94	0.08 (0.07–0.09)
Metric	675.58** (304)	0.93	0.94	0.09 (0.08–0.10)	−248,21	−104	0.03	0.01
Scalar	480.72** (245)	0.96	0.95	0.08 (0.07–0.09)	−194,86	59	−0.02	0.01
Age group
Configural	477.92** (200)	0.95	0.93	0.09 (0.08–0.10)
Metric	530.11** (234)	0.94	0.94	0.09 (0.08–0.10)	−52.19	−34	0.01	0.00
Scalar	528.12** (245)	0.95	0.94	0.09 (0.08–0.10)	1.99	−11	−0.01	0.00

** = p < .001.χ2 = Chi square; df = degrees of freedom; CFI = Comparative Fit Index; RMSEA = Root Mean Square Error of Approximation; CI = Confidence Interval; TLI = Tucker-Lewis Index; ΔCFI = Change in Comparative Fix Index. For gender groups (girls and boys) Satorra-Bentler $\mathrm{\Delta }\mathit{\chi }$²: Config vs. Metric $\mathit{\chi }$² (104) = 171.59, p < 0.001. Metric vs. Scaler $\mathit{\chi }$² (59) = 103.93, p < 0.001. For age groups: Satorra-Bentler $\mathrm{\Delta }\mathit{\chi }$²: Config vs. Metric $\mathit{\chi }$² (34) = 29.26, p = 0.699. Metric vs. Scaler $\mathit{\chi }$² (11) = 6.04, p = 0.871.

Measurement invariance for age

Two age groups were used for the calculations, namely 15–17 (n = 205, M = 16,50, SD =.56) and 18–20 (n = 111, M = 18.27, SD =.54). The groups were divided according to Sweden’s border between child and adult (>18 years).

The configural invariance model demonstrated an excellent fit to the data, indicating that the factor structure of SCS-Y-SE was equivalent for the two age groups, the metric invariance model showed an excellent fit, indicating that the factor loadings of the items were the same between the age groups; the scalar invariance model showed an excellent fit, indicating that there are no observable age differences in the intercepts (Table 6).

Differences between gender, age groups and socioeconomic status

Independent t-tests between girls and boys showed that girls scored lower on the total self-compassion scale than boys (t(314) = −3.32, p < .001) with a small effect size (d = −0.40). Girls scored higher on the following subscales than boys: self-judgement (t(314) = −3.87, p < .001; small effect size), isolation (t(314) = −3.73, p < .001; small effect size) and over-identification (t(314) = −4.47, p < .001; medium effect size) (Table 1).

We tested age-differences on the total scale of SCS-Y-SE with a t-test (group 1, 15–17 (n = 205, M = 2,99, SD = .66); group 2, 18–20 (n = 111, M = 3.03, SD = .73)) but found no significant differences (t (314) = −.34, p = .735). Socioeconomic impact on self-compassion was tested with Pearson’s correlation, and no correlation was found between SCS-Y-SE (total) and SES (r = −.08, p = .130).

Discussion

The Swedish version SCS-Y-SE was shown to be a reliable self-report questionnaire (with some limitations) in line with the validation results of the United States’ version of the scale (Neff et al., 2021). SCS-Y-SE had good internal consistency overall, the exception being the subscale mindfulness for boys (as was also seen in Neff et al. (2021)). SCS-Y-SE showed moderate to good test-retest reliability and ICC (Koo & Li, 2016), over a period of three weeks (.66–.74 for the subscales and .86 for the total scale), similar to SCS-Y’s score (.83) for the same test-retest period.

Differences related to sex and age groups

We found that girls had more self-judgement, more sense of isolation and more over-identification than boys, which is in line with previous adolescent (Bluth & Blanton, 2015; Bluth et al., 2017; Henje et al., 2020) and adult (Yarnell et al., 2015) compassion research. There can be several complex interacting explanations for this pattern, including biological (Susman & Dorn, 2009), social (Connell, 2009) and psychological factors (Susman & Dorn, 2009). For example, girls are generally more sensitive to relational problems because they are raised to be more relational than boys (Connell, 2009). Another explanation could be the self-objectification and body shame that many girls suffer from (Liss & Erchull, 2015). Similar to Neff et al. (2021), we found no age differences in self-compassion.

Confirmatory factor analyses

Both the single-bifactor model and the two-bifactor model using CFA had a good fit in SCS-Y-SE.

How to best measure the construct of self-compassion has been heavily debated (Cleare et al., 2018; Ferrari et al., 2022; Muris & Petrocchi, 2017). Neff et al. (2019) compared various models in 20 international samples of adults and found that fit was excellent using ESEM for the 6-factor correlated, single-bifactor and correlated two-bifactor models and that for SCS-Y in particular, a single-bifactor ESEM had the best fit. Our best models were the single-bifactor model and the two-bifactor model (both using CFA) without any significant differences. According to the principle of parsimony, and in line with Neff et al. (2019), we propose that the single-bifactor model should be used. The question of whether CSR and UCS are one single dimension or two semi-independent, unipolar dimensions lies outside the scope of this research.

Measurement invariance

We found measurement invariance for age and ambiguous invariance for sex. For age, the factor structure, loading and intercept were the same. For sex, the factor structure was the same between groups according to the configural invariance model, but metric inequality indicated that one or more items of the questionnaire resulted in response bias (i.e., were answered differently by girls and boys). In addition, scalar invariance for sex was not found, and the intercepts were interpreted as inequal. We conclude that the test can be used across the tested ages (15–20), but it is unclear whether it can be used equally well by both sexes.

Convergent and divergent validity

Evidence was found for convergent and divergent validity using correlations with other self-report questionnaires of similar and different constructs. These findings are in line with previous findings that self-reported self-compassion negatively correlates with anxiety (Karen Bluth, Gaylord, et al., 2016) (Neff & McGehee, 2010), depression (Henje et al., 2020; Neff & McGehee, 2010) and anger (Henje et al., 2020), and positively correlates with well-being (Bluth & Blanton, 2015; Bluth et al., 2017; Henje et al., 2020). Self-compassion had a much lower correlation with other constructs like the SDQ impact scale and pain interference as measured by the PROMIS-instrument.

Limitations

Participants were not geographically stratified, and a convenience sample was used, which limits generalisability. SCS-Y-SE was validated for the age span 15 to 20 years, and these results are therefore not directly comparable to the original sample (ages 10–14) (Neff et al., 2021). Another limitation was that several of these questionnaires were not validated for ages 19–20. More girls than boys participated in the study (67%). Measurement invariance calculations were done even though the sample sizes were quite small (100 to 150 in each group), since good results have been reported with both small and skewed samples (Cheung & Rensvold, 2002; Finch et al., 2018).

Conclusions

The single bi-factor model with six subscales using CFA had a good fit. We recommend further research to ensure measurement invariance for sex. The instrument could be interpreted over the included ages (15–20), but not the sexes. SCS-Y is therefore validated (with some limitations) in our Swedish school sample.

In addition, we found that CSR and UCS could be both one single dimension or two semi-independent unipolar dimensions. However, we recommend following the principle of parsimony and Neff et al. (2021) original theory that one single dimension of self-compassion is the best model for the scale. The Swedish version of SCS-Y (SCS-Y-SE) will be useful in future Swedish self-compassion research and for measuring self-compassion in therapy. SCS-Y-SE can be used for measuring self-compassion as a protective factor.

Open Scholarship

This article has earned the Center for Open Science badge for Preregistered. The materials are openly accessible at https://doi.org/10.1080/00050067.2023.2217326

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author ID upon reasonable request.

Supplementary material

Supplemental data for this article can be accessed at https://doi.org/10.1080/00050067.2023.2217326.

Additional information

Funding

The work was supported by the Region Västerbotten (ALF) Clinical Science, Umeå University.