Measuring self-compassion in people living with dementia: investigating the validity of the Self-Compassion Scale-Short form (SCS-SF)

Abstract

Objectives

Self-compassion may be a psychological resource for living well with dementia, but research is limited by the lack of a validated self-compassion measure for people with dementia. This study aimed to explore the SCS-SF’s psychometric properties as well as correlates of self-compassion for people with dementia.

Method

A total of 207 people with dementia were recruited to a cross-sectional survey involving the SCS-SF and measures of well-being, self-esteem, and depression. Data analyses (n = 193) included internal consistency reliability, correlational analyses, Exploratory Factor Analysis (EFA), plus ANOVAs and t-tests.

Results

Self-compassion significantly correlated positively with well-being and self-esteem, and negatively with depression. Reliability and preliminary construct validity of the SCS-SF was supported. EFA suggested two underlying factors formed by positive and negative components of self-compassion. The negative factor explained more variance and showed stronger correlations with total self-compassion, well-being, self-esteem, and depression compared to the positive factor. Self-compassion significantly differed based on age but not gender, dementia subtype or time since diagnosis.

Conclusion

The SCS-SF shows potential as a valid and reliable measure of self-compassion for people with dementia, but further research is needed. The SCS-SF may measure two distinct constructs, which possibly play different roles in relation to well-being in dementia: self-compassion and self-criticism. Clinicians and researchers may wish to interpret these factors separately.

Keywords:

• Dementia
• measure
• self-compassion
• reliability
• validity

Introduction

Compassion flows in three directions; towards others, from others and towards the self (Gilbert, 2009). Self-compassion can be understood as a positive attitude of kindness and non-judgement towards the self (Neff, 2003b), and has been associated with well-being across a range of populations (Zessin et al., 2015), including older adults (Brown et al., 2019; Tavares et al., 2023). Neff (2003b) conceptualises self-compassion as consisting of three main elements: self-kindness (being kind and understanding towards oneself), common humanity (recognising that suffering is part of the human shared experience), and mindfulness (observing thoughts and feelings for what they are).

Research with older adults in non-clinical groups suggests self-compassion is involved in positive adjustment in ageing (Allen & Leary, 2013; Phillips & Ferguson, 2013), potentially in relation to life-review and acceptance (Erikson & Erikson, 1998; Neff, 2003b). Older men may report greater self-compassion compared to females (Bratt & Fagerström, 2020) and self-compassion has been found to increase with age (Homan, 2016). It may be that self-compassion is predictive of ego integrity (Erikson & Erikson, 1998) and moderated by age, i.e. older adults with high ego integrity are able to be kinder towards themselves and accepting of past adverse experiences (Phillips & Ferguson, 2013). Whilst clear evidence supporting these relationships is currently equivocal, research does indicate that interventions which promote self-compassion may benefit well-being in older populations (e.g. Allen et al., 2012; Homan, 2016; Kim & Ko, 2018).

Insofar as self-compassion may contribute to adaptive coping and reduced stress whilst living with adversity such as chronic illness (Sirois et al., 2015), it is likely to contribute to maintaining well-being in later life during adversity and therefore may be of value for older people living with dementia. Research has shown that other psychological resources relating to the self, such as self-esteem and self-efficacy are associated with the capacity to live well with dementia and could therefore be targets for interventions (Lamont et al., 2019). However, to date, research on self-compassion and dementia is extremely limited.

Whilst interventions that include components of self-compassion, such as mindfulness-based interventions (Berk et al., 2018; Hoffman et al., 2020) and Compassion Focussed Therapy (Collins et al., 2018), have demonstrated positive outcomes relating to well-being in dementia, the role of self-compassion remains unclear since little research has specifically measured self-compassion in dementia. Craig et al. (2018) found improvements in self-compassion mood and anxiety in dementia following Compassion Focussed Therapy and concluded that self-compassion may be a protective factor as dementia progresses. On the other hand, Berk et al. (2019) found no evidence that a mindfulness intervention led to improvements in self-compassion or quality of life for people with dementia and self-compassion did not mediate improvements in mood following a meditation programme evaluated by Innes et al. (2012), although participants in this study also included caregivers and people with mild cognitive impairment.

It is possible that mixed findings regarding the role of self-compassion in improving well-being in dementia reflects measurement error, as a major limitation of existing studies (Berk et al., 2019; Craig et al., 2018; Innes et al., 2012) is the lack of a self-compassion measure that has been validated and is acceptable for people with dementia. Innes et al. (2012) used the 26 item Self-compassion Scale [SCS]; Neff, 2003a), whereas Berk et al. (2019) and Craig et al. (2018) used the 12-item adaptation of this scale, the SCS-SF (Raes et al., 2011). Both the SCS and SCS-SF have demonstrated good psychometric properties for different groups (Neff, 2003a; Raes et al., 2011), however, there remains a need to evaluate whether this equally applies for people living with dementia experiencing varying levels of cognitive impairment (Clarke et al., 2020; Schölzel‐Dorenbos et al., 2007). It is currently unclear whether the SCS-SF is a valid and reliable measure of self-compassion for people with dementia.

Specifically, the construct validity and factor structure of the SCS-SF in dementia has yet to be investigated. Within other populations, a first order factor (self-compassion) with six second order factors based on the subscales self-kindness (SK), common humanity (CG), mindfulness (MI), self-judgement (SJ), isolation (IS) and overidentification (OI) has been supported for the SCS-SF (Castilho et al., 2015; Raes et al., 2011). Uršič et al. (2019) found a six-factor model based on the six subscales but were unable to replicate the higher order factor. A bifactor model (Rocha et al., 2022), a two factor (positive and negative factor) model (Babenko & Guo, 2019; Bratt & Fagerström, 2020; Hayes et al., 2016; Kotera & Sheffield, 2020; Lluch-Sanz et al., 2022) and a three-factor model (one positive, two negative factors) with 10 items (Meng et al., 2019) have shown a better fit across a range of populations with English or translated SCS-SF versions. However, Bratt and Fagerström (2020) were unable to confirm the proposed two factor structure for older adults in Sweden. Given these equivocal findings, the factor structure and construct validity of the SCS-SF in dementia awaits investigation.

Research aims

This study aimed to explore how the SCS-SF performs for people with dementia by investigating the validity and reliability of the scale. Exploring the underlying dimensions of the SCS-SF, the association between self-compassion and well-being, and whether self-compassion differs based on demographic variables in dementia were specific aims. In relation to the aims, the study sought to answer the following questions:

Primary question

• Is the SCS-SF a reliable and valid measure of self-compassion for people living with dementia?

Secondary questions

• What is the factor structure of the SCS-SF for people with dementia?

• Does self-compassion measured using the SCS-SF correlate with well-being for people with dementia?

• Does self-compassion in people with dementia, measured using the SCS-SF, differ based on gender, age, dementia subtype or time since diagnosis?

Regarding validity (primary question) and the correlation between self-compassion and well-being (secondary question); based on previous research (Homan, 2016; Hwang et al., 2016) the study sought to test the hypothesis (H1) that self-compassion (SCS-SF) would positively correlate with measures of well-being and self-esteem, and negatively correlate with a measure of depression. Hypotheses were not proposed for the remaining research questions due to the exploratory aims.

Materials and methods

Design and participants

A quantitative cross-sectional, survey-based design was utilised with a volunteer sample of people with dementia living in the community. To enhance accessibility, prior to recruitment, the study advertisement poster and participant information documents were reviewed by people living with dementia at a Patient and Public Involvement (PPI) group and changes to the documents to improve accessibility were made. Participants were recruited from Join Dementia Research (JDR; an online UK dementia research registry), as well as regional and national dementia charities via posters, social media, and word of mouth. Two National Health Service (NHS) Trusts in the UK also supported study promotion and recruitment amongst people with dementia using their services.

Eligible participants had to have a diagnosis of dementia (self-reported) and able to read English. To remain inclusive and maximise recruitment this included any subtype of dementia, any length of time since diagnosis and all ages. No further exclusion criteria were applied.

Procedure

Recruitment ran between October 2022 and March 2023. Participants accessed the measures online (Jisc Online Surveys) via the QR code or direct link within the advert. Upon opening the link, a short summary of the study information was presented, and participants were advised to follow a link to access the full information document. Participants then proceeded to a consent page before being able to access the measures. On completion, participants were able to opt in if they wanted to be contacted by the researcher (via email) to hear about the study results. Finally, participants were presented with a debrief page and sources for support.

A paper option was available by request and was posted to the participant’s home address, along with an information sheet, debrief sheet and a pre-paid envelope to return.

Measures

Self-compassion

The 12-item Self-compassion Scale (SCS-SF; Raes et al., 2011) asks participants to rate how often they behave in the manner stated in each item using a five-point scale (almost never to almost always). The scale contains the subscales self-kindness (SK), mindfulness (MI), common humanity (CH), self-judgement (SJ), isolation (IS), over-identified (OI). Negative items (1, 4, 8, 9, 11, 12) are reversed scored and, since development, the authors recommend calculating the total mean SCS-SF rather than the sum (Neff, n.d.), therefore total mean is used within the present study. Due to low Cronbach’s alpha for the SCS-SF subscales individually, Raes et al. (2011) recommend using the total score, which they demonstrated has validity and adequate internal consistency reliability (Cronbach’s alpha ≥0.86). Scores range from 1 to 5 and high scores indicate greater self-compassion.

Well-being

The adapted 12-item Control, Autonomy, Self-realisation, and Pleasure Scale (CASP-12 v2; Wiggins et al., 2008) was utilised (with permission). This instrument is based on the original 19-item version (CASP-19; Hyde et al., 2003) but shows stronger psychometric properties and may be more robust for people with dementia (Stoner et al., 2019). Unlike previous shortened versions (Börsch-Supan et al., 2005), the CASP-12 v2 combines the control and autonomy subscales into one subscale. Participants rate the extent that items apply to them using a four-point scale (often to never), four items are reverse scored, and a higher total score indicates greater well-being (scores range from 12 to 48). Whilst the scale is defined as a quality of life (QoL) measure, previous research has used it to measure well-being (Okely et al., 2016), since the two constructs have been viewed interchangeably. The scale has demonstrated an internal reliability of α = 0.82 for people over the age of 50 (Okely et al., 2016). In the present study Cronbach’s alpha for the CASP-12 v2 was α = 0.830.

Self-esteem

The Rosenberg Self-esteem Scale (RSES; Rosenberg, 1965) was utilised to measure participants’ beliefs and attitudes towards themselves across 10 items using a five-point scale. Five items are reverse scored and higher total scores indicate greater self-esteem (scores range from 10 to 40). The RSES has demonstrated good internal reliability (α = 0.82 and α = 0.83) and face validity in dementia (Burgener & Berger, 2008; Lamont et al., 2019) and has been used in research to examine construct validity for a self-stigma scale in dementia (Burgener & Berger, 2008). In the present study Cronbach’s alpha for the RSES was α = 0.884.

Depression

The 15-item item Geriatric Depression Scale (GDS-15; Sheikh & Yesavage, 1986) was used to assess symptoms of depression. Participants respond ‘yes’ or ‘no’ to items; ratings that are indicative of depression score one point. Scores range from 0 to 15 and a higher total score suggests greater levels of depression. The scale has demonstrated good reliability (α = 0.87; Lach et al., 2010) and validity for people with dementia (e.g. Burgener & Berger, 2008; Lach et al., 2010). In the present study Cronbach’s alpha for the GDS-15 was α = 0.871.

Demographics

Participants were asked to self-report their age, gender, subtype of dementia and time since diagnosis.

Data analysis

Data were analysed using IBM SPSS Statistics 28.0. Convergent and discriminant validity of the SCS-SF was analysed using Pearson’s correlational analysis for the SCS-SF correlations with the RSES, GDS-15 and CASP-12 v2. Internal consistency reliabilities for each measure were calculated. Power analysis for correlations between the SCS-SF and measures of well-being, self-esteem, and depression, based on correlations within existing literature (Hwang et al., 2016) and common validity cut offs (Abma et al., 2016; Cohen, 1988), suggested a priori minimum sample size of 139.

An Exploratory Factor Analysis (EFA) using Principal Axis Factoring (PAF) and oblique rotation for the SCS-SF was conducted to determine the possible underlying dimensions of the scale for people with dementia and to further assess construct validity. EFA was chosen, rather than CFA, since the differing factor structures reported in regard to other populations mean it is unclear what the factor structure may be for people with dementia and there were therefore insufficient grounds to hypothesise a specific factor structure. PAF was chosen, as opposed to alternate EFA extraction methods such as Maximum Likelihood (ML) and Weighted Least Squares (WLS), as it has been highlighted (e.g. Watkins, 2018) that PAF outperforms ML when the sample size is small (<300) and the relationship between some items and factors are weak (≤0.40); conditions present within the current study. WLS has been suggested to require a strict large sample size as an EFA method and in addition, whilst other well-established extraction methods (e.g. Weighted Least Squares Mean and Variance adjusted) are recommended when carrying out CFA, PAF is typically judged more applicable to EFA and is less prone to improper solutions than ML whilst also being less reliant on normality assumptions holding (e.g. Kyriazos & Poga-Kyriazou, 2023) so was used in preference for this study. Oblique rotation was selected as this allows factors to be correlated (Field, 2018) and previous research suggests that correlations between self-compassion factors and variables are likely (Raes et al., 2011). Whilst there are no strict rules for EFA sample size, the prevalent 10:1 participant to item ratio rule-of-thumb (Costello & Osborne, 2005) suggested a priori minimum sample size of 120.

In addition, ANOVA and t-tests were conducted to determine whether there were any differences in self-compassion based on age, gender, dementia subtype or time since diagnosis.

Ethics

All participant data was anonymised as no identifiable data was collected within the measures and random numbers automatically assigned to each data set as they were submitted. The information sheet informed participants they had the right to withdraw their informed consent and data up until submitting their responses as after this point anonymised data were pooled.

Results

A total of 207 participants attempted the measures; 179 of these accessed the measures online. However, 14 participants’ data sets were excluded (6.76%) due to missing data on the SCS-SF, therefore 193 participant data sets were analysed. Of these 193 data sets, two participants completed <50% of the CASP-12 v2 and one participant completed <50% of the GDS-15, therefore these three participants were excluded from analyses. Of the remaining participant data sets, seven responses were missing on the CASP-12 v2, 12 on the RSES and 11 on the GDS-15. No single participant missed more than two items in any one measure. Little’s (1988) Missing Completely at Random (MCAR) test was non-significant (>0.05); therefore, the data were assumed to be MCAR. These 30 missing responses were replaced using Multiple Imputation (see Schafer & Graham, 2002).

Skew and kurtosis of participant data for each measure, t-test and ANOVA group were reviewed (see Supplementary file 1) and except for the time since diagnosis group ‘6 months − 1 year’ (flagged as leptokurtic, kurtosis 2.1), all values were in the acceptable range of −2 to 2 (George & Mallery, 2022). No extreme outliers (>2 standard deviations away from the mean) were identified from box plots. Assuming central limit theorem (Field, 2018); normality tests were not utilised for the measures/t-tests but were for the ANOVAs due to smaller group sizes. Shapiro-Wilks test only showed a significant departure from normality for ‘5-10 years’; W(41) = 0.926, p = 0.01. Homogeneity of variances was demonstrated by Levene’s test for gender (p = 0.509), age (p = 0.950), and diagnosis subtype (p = 0.704). Levene’s test rejected the null hypothesis of equal variances for time since diagnosis, F(4, 187) = 3.50, p = .009. Considering this and the normality violation, the non-parametric Kruskal-Wallis test was conducted for time since diagnosis.

Participant demographic information is displayed in Table 1. A total of 51 participants (26.4%) reported receiving help to complete the measures from their partner (41), child (5), grandchild (1), care worker (1) or the researcher (3).

Table 1. Summary of participant demographics, total mean score, and standard deviation (SD) on the SCS-SF.

Demographic	Number of participants (%)	Total mean SCS-SF score (SD)
Gender
Male	119 (61.7)	3.21 (.758)
Female	73 (37.8)	3.06 (.703)
Missing	1 (0.52)	3.33
Age
50–64 years old	42 (21.8)	2.83 (.731)
65 or over	150 (77.7)	3.25 (.717)
Missing	1 (0.52)	3.08
Diagnosis subtype
Alzheimer’s dementia	108 (60.0)	3.25 (.755)
Vascular dementia	20 (10.4)	2.95 (.624)
Mixed dementia	31 (16.1)	3.02 (.754)
Mixed Alzheimer’s and Vascular	10	2.825 (.872)
Other mixed dementia	6	2.83 (.447)
Mixed subtypes not reported	15	3.23 (.753)
Frontotemporal dementia (FTD)	19 (9.84)	3.17 (.707)
Dementia with Lewy Bodies (DLB)	8 (4.15)	3.18 (.884)
Posterior Cortical Atrophy (PCA)	2 (1.04)	2.17 (.236)
Not known	5 (2.59)	3.17 (.257)
Time since diagnosis
Less than 6 months	19 (9.84)	3.11 (.502)
6 months − 1 year	31 (16.1)	3.11 (.635)
1–5 years	93 (48.2)	3.16 (.792)
5–10 years	41 (21.2)	3.21 (.830)
Over 10 years	8 (4.15)	3.29 (.452)
Not known	1 (0.52)	2.25

Reliability analysis

The internal consistency reliability for the total mean SCS-SF (n = 193) was α = 0.796 (mean =3.16, standard deviation = 0.738). Only the removal of item 10 (when I feel inadequate in some way, I try to remind myself that feelings of inadequacy are shared by most people) resulted in an increase in Cronbach’s alpha for the total SCS-SF score, but this change was marginal (from 0.796 to 0.800) and therefore no items were removed from the measure.

Exploratory factor analysis

Suitability for EFA of the SCS-SF was confirmed as multicollinearity screening showed that the R-matrix (0.022) determinant was greater than 0.00001 and inter-item correlations were below the cut off r = 0.9 (Field, 2018). All items had some correlations that were above the recommend minimum correlation r = 0.3 (Field, 2018; see Supplementary file 2). Also, Bartlett’s test of sphericity was significant (χ² (66) = 714.846, p < 0.001) and Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.819, which is considered ‘meritorious’ for EFA (Kaiser & Rice, 1974). All KMO values for individual items were >0.7; therefore, above the acceptable limit of 0.5 (Kaiser & Rice, 1974). Parallel analysis (O’Connor, 2000) for PAF (as recommended by Wood et al., 2015) indicated that a two or three factor model would be the best fit (see Supplementary file 3).

A preliminary analysis to obtain eigenvalues for factors revealed a potential three-factor model (see Supplementary file 4). Three factors had eigenvalues over Kaiser’s criterion of 1, although factor 3’s eigenvalue (1.012) just met this cut off (Field, 2018). In combination, the three factors explained 59.7% of the variance. Two items loaded onto factor 3 (item 6 and 10) and, except for item 10 (0.423), all factor loadings in the pattern matrix were >0.45 which is considered an acceptable factor loading for sample sizes above 150 (Hair et al., 2010). The scree plot (see Supplementary file 5) demonstrated inflexions that would justify retaining two or three factors. However, the structure matrix (item and factor correlations), showed items cross-loading between factor 2 and 3. Extraction communalities were low to moderate, between 0.40 and 0.70 (Costello & Osborne, 2005), for eight items and <0.40 for four items. Factors with less than three items may be unstable (Costello & Osborne, 2005). As such, since the scree plot showed the potential for two factors, there was heavy cross-loading, and factor 3’s eigenvalue was relatively low, a two-factor solution was explored.

The subsequent two-factor model explained 51.4% of the variance (31.8% factor 1, 19.6% factor 2) and only item 6 cross-loaded. Table 2 presents the pattern and structure coefficients after rotation. All item loadings were above the cut off of 0.3 (Costello & Osborne, 2005). Similar to the three-factor model, extraction communalities for eight items were low to moderate, between 0.40 and 0.70 (Costello & Osborne, 2005). For four items, extraction communalities were <0.40 thus indicating that the proportion of each item’s variance which can be explained by the factors is limited. The two-factor model was retained because aside from item 6, the items that clustered on the same factors suggested that factor 1 represents negative indicators of self-compassion or a lack of self-compassion as consists of negatively worded items (e.g. 1, 4, 8, 9, 11, 12) and factor 2 represents positive indicators or the presence of self-compassion as consists of the positively worded items, with minimal cross loadings. Cronbach’s alpha was calculated for both factors and did not improve if any items were deleted.

Table 2. EFA summary of the two-factor solution for the SCS-SF.

Item	Rotated factor loadings (pattern coefficients)		Structure coefficients
	Factor 1	Factor 2	Factor 1	Factor 2
11. I’m disapproving and judgmental about my own flaws and inadequacies.	.838		.820
8. When I fail at something that’s important to me, I tend to feel alone in my failure.	.698		.671
12. I’m intolerant and impatient towards those aspects of my personality I don’t like.	.663		.669
1. When I fail at something important to me I become consumed by feelings of inadequacy.	.654		.675
9. When I’m feeling down I tend to obsess and fixate on everything that’s wrong.	.648		.647
4. When I’m feeling down, I tend to feel like most other people are probably happier than I am.	.614		.620
3. When something painful happens I try to take a balanced view of the situation.		.769		.763
2. I try to be understanding and patient towards those aspects of my personality I don’t like.		.707		.686
5. I try to see my failings as part of the human condition.		.660		.644
7. When something upsets me I try to keep my emotions in balance.		.535		.559
10. When I feel inadequate in some way, I try to remind myself that feelings of inadequacy are shared by most people.		.371		.373
6. When I’m going through a very hard time, I give myself the caring and tenderness I need.	.322	.362	.409	.439
Eigenvalues	3.82	2.35
% of variance	31.8	19.6
Cronbach’s alpha reliability	.839	.746

Note. n = 193, only coefficients above 0.3 are shown. For Cronbach’s alpha reliability Factor 1 included items 1, 4, 8, 9, 11, 12. Factor 2 included items 2, 3, 5, 6, 7, 10.

Correlational analyses

Table 3 presents correlations between the CASP-12 v2, RSES and GDS-15 with the SCS-SF total mean and the positive (M = 3.20, SD = 0.838) and negative (M = 3.11, SD = 1.04) factor scores. Factor 1 (negative factor) included items 1, 4, 8, 9, 11, 12, whilst Factor 2 (positive factor) included items 2, 3, 5, 6, 7, 10. The SCS-SF (total score) demonstrated significant positive correlations with the CASP-12 v2 and RSES, and significant negative correlations with the GDS-15, with large (>0.5) effect sizes (Cohen, 1988).

Table 3. Correlations between SCS-SF total mean (plus sub-factors identified through EFA) and CASP-12 v2, RSES, and the GDS-15 total scores.

Scales	SCS-SF Total Mean	SCS-SF Factor 1	SCS-SF Factor 2
CASP-12 v2^a	.550**	.512**	.341**
Control and Autonomy subscale	.522**	.486**	.322**
Pleasure subscale	.425**	.404**	.252**
Self-reaslisation subscale	.421**	.382**	.273**
RSES^b	.680**	.660**	.381**
GDS-15^c	−0.541**	−0.554**	−0.268**
SCS-SF Total Mean^b		.833**	.730**
SCS-SF Factor 1^b			.230*

** p < 0.001.

* p < 0.01 Sample size.

^a n = 191.

^b n = 193.

^c n = 192.

CASP-12 v2: Control, Autonomy, Self-realisation, and Pleasure Scale.

RSES: Rosenberg Self-esteem Scale.

GDS-15: Geriatric Depression Scale.

Demographic group differences in self-compassion

Results from the independent t-tests, one-way ANOVA and Kruskal-Wallis for group differences in total mean SCS-SF (Table 4) showed that those over 65 years old reported significantly greater levels of self-compassion compared to those aged 50–64, with a large (>0.5) effect size (Cohen, 1988).

Table 4. ANOVA, Kruskal-Wallis, and t-test statistics for demographic group differences in total mean SCS-SF.

Demographic group	Test	P value	Cohens d	Eta-squared
Gender^a (t-test; male, female)	t = 1.41	.161	.203
Age^a (t-test; 50–64 years old, 65 or over)	t = 3.37	<.001*	.569
Diagnosis subtype^b (ANOVA; Alzheimer’s, Vascular, Mixed dementia, ‘Other’ dementia’s)	F = 1.42	.239		.0226
Time since diagnosis^a (Kruskal-Wallis; less than 6 months, 6 months-1 year, 1–5 years, 5–10 years, over 10 years)	H=.866	.929		.0275

* p < 0.001, Sample size.

^a 192.

^b 188, Diagnosis subtype; ‘Other’ dementia group comprised FTD, DLB and PCA.

Discussion

The absence of a validated measure of self-compassion in dementia has been a major limitation of existing studies and a barrier for further research. This is the first study to address this by exploring the psychometric properties of the SCS-SF, as well as correlates of self-compassion, amongst people living with dementia. Aligned with the aims of this study, results showed evidence that the SCS-SF has a two-factor structure amongst people with dementia and that self-compassion is associated with aspects of well-being and may differ according to age. The results of this study may therefore inform the development and evaluation of interventions that are most appropriate and beneficial for supporting the well-being of people with dementia.

Correlational analyses between the total mean SCS-SF with the CASP-12 v2, RSES and GDS-15 supports the convergent validity of the SCS-SF for people with dementia and the proposed hypothesis that self-compassion would correlate positively with well-being and self-esteem, and negatively with depression. These findings are consistent with existing literature involving older adults (e.g. Homan, 2016) and other populations (e.g. Hwang et al., 2016; Neff, 2003a). However, future research is needed to clarify the causal direction of these relationships in dementia. Given that treating oneself with kindness and common humanity are components of self-compassion (Neff, 2003b), it may be that self-compassion helps to maintain sense of identity and connection with others (Homan, 2016). Therefore, if self-compassion facilitates important psychological needs in dementia such as identity and inclusion (Kitwood, 1997), future research should consider the potential role of self-compassion as a mediator or moderator (e.g. Tavares et al., 2023) for improving well-being in dementia and, considering the currently equivocal literature (e.g. Berk et al., 2019; Craig et al., 2018; Innes et al., 2012), how this may differ across interventions.

EFA indicated that the underlying dimensions of the SCS-SF consisted of a positive and negative factor that explained 51.4% of variance. This finding is consistent with several other studies which have confirmed a similar two factor structure amongst students (Kotera & Sheffield, 2020), nurses (Lluch-Sanz et al., 2022) and psychotherapy clients (Hayes et al., 2016). In addition, these studies found the two factors explained similar levels of variance to that found in the present study, including the negative factor explaining more variance than the positive factor. Whilst a similar two factor structure, which explained 39.7% of variance, was initially demonstrated for older adults; CFA found the two-factor structure was not an acceptable fit (Bratt & Fagerström, 2020). Considering this, there may be an alternative factor structure which explains more variance but was not detected based on the existing scale and current sample, therefore further research (e.g. CFA) is needed to confirm this factor structure (and variance explained) for people with dementia.

The two factors demonstrated some evidence of discriminant validity due to minimal cross-loading, with the exception of items 6 (when I’m going through a very hard time, I give myself the caring and tenderness I need) and 10 (when I feel inadequate in some way, I try to remind myself that feelings of inadequacy are shared by most people). These items also demonstrated cross-loading and/or low factor loadings according to Bratt and Fagerström (2020) using a Swedish translation of the SCS-SF amongst older people. This suggests that, although these items did not reduce the reliability of the SCS-SF for people with dementia and older adults, these items may not measure self-compassion as accurately as for other populations/age groups. Alternatively, the longer length of these items may have impacted accessibility for people with dementia. Future research with older people and/or people with dementia could focus on adapting these items and testing item-level psychometric properties.

Unlike previous research in which up to 71% of participants with dementia required help to complete the SCS-SF (Berk et al., 2019), only 26.4% of participants in the present study reported receiving help with reading and/or recording their responses to the set of measures; suggesting that the SCS-SF was accessible to the majority as a self-report instrument. The online format may have resulted in inclusion of more participants who were able to independently complete the measure, contributing to this disparity. However, it is also not possible to confirm that, no participant received excessive help. Future qualitative research may be of benefit to explore the accessibility of the SCS-SF for people with dementia and potential barriers to this. The weak (<0.3) positive correlation (Cohen, 1988) found between the negative (reverse scored) and positive factor contradicts suggestions that they may represent opposing poles of self-compassion (Lluch-Sanz et al., 2022) as a strong correlation would be expected. Instead, the negative factor may represent self-criticism (Hayes et al., 2016; Kotera & Sheffield, 2020; López et al., 2015) which is inversely associated with self-compassion (Zhang et al., 2019) and posited as a distinct construct from self-compassion (Neff, 2011, p. 165) with different neural activity (Lutz et al., 2020). As well as explaining more variance in self-compassion, the negative factor (higher scores indicative of lower self-criticism) showed relatively stronger correlations with total self-compassion, well-being, self-esteem, and depression compared to the positive factor (self-compassion), suggesting that individuals with lower levels of self-criticism may also experience greater well-being and self-esteem as well as lower levels of depression, in addition to supporting the distinction between the factors. Considering this, whilst total SCS-SF scores showed acceptable internal consistency reliability (George & Mallery, 2022), conceptually it may be more appropriate to measure the two factors separately for people with dementia as they showed acceptable (positive factor) and good (negative factor) internal consistency reliability separately (George & Mallery, 2022).

A pertinent implication from these findings is that for people with dementia the relative absence of self-criticism may be more important for well-being than the presence of self-compassion. It may be that people with dementia find self-compassion more difficult, possibly moderated by ageing, or due to the impact of negativity bias, which may be heightened in dementia (Fleming et al., 2003) and are therefore more likely to attend to and be impacted by changes in self-criticism relative to self-compassion. Furthermore, self-compassion may originate in care seeking/giving behaviours developed from childhood (Gilbert, 2009) and therefore may be a more stable self to self relating process over the lifespan, compared to self-criticism, which is influenced by social and environmental factors daily (Veilleux et al., 2024). Further research is needed to investigate the predictive capabilities of self-compassion and self-criticism for well-being in dementia.

The potential stability of self-compassion may also explain the finding that SCS-SF scores did not differ depending on time since diagnosis. Alternatively, it may be that self-compassion differs between stages of dementia as people navigate challenges relating to expectations and reality of dementia progression (e.g. Read et al., 2017). Considering variation in dementia progression and delays in diagnosis and help-seeking (Parker et al., 2020), time since diagnosis may not be an accurate reflection of a person’s stage of dementia or cognitive ability. Cognitive impairment or stage of dementia was not measured in this study, and this is therefore a limitation in terms of potential generalisability which future research could address. Consistent with Phillips and Ferguson (2013), no differences in gender were found, suggesting that the SCS-SF measures self-compassion similarly for men and women. This was the first study to explore differences in self-compassion based on dementia subtype; whilst no differences were found, this should be explored with a larger sample.

The significant difference in self-compassion for participants living with dementia aged 50 to 64 compared to over 65 is consistent with previous findings that self-compassion increases with age over the lifespan (Homan, 2016; Hwang et al., 2016). Therefore, if self-compassion is involved in establishing ego integrity in older age (Erikson & Erikson, 1998; Phillips & Ferguson, 2013), it is possible that dementia does not necessarily impact this process. The association between self-criticism and age over time in dementia may be different and potentially be confounded by bio-psycho-social factors (e.g. multi-morbidity or living alone) and future research could investigate this.

Limitations

Several limitations should be considered. Firstly, whilst the a priori sample size target was achieved, a larger sample (>300) may have provided a more stable EFA solution (Field, 2018). In addition, whilst our approach of combining Parallel Analysis with PAF improves robustness of factor selection in EFA (Wood et al., 2015), there are more robust methods of estimation compared to PAF, such as WLS and WLSMV, that were not possible to use within the current study since they are better suited to CFA. Therefore, future studies aiming to confirm the factor structure of the SCS-SF in dementia should seek to carry out CFA and in doing so maximise the robustness of factor extraction methods to increase the validity and replicability of our proposed factor structure. Secondly, the opportunistic sampling of participants combined with use of EFA admittedly places limits on the generalisability of our findings; whilst study promotion and inclusion criteria aimed at inclusivity, individuals outside of the regional area with limited access to technology may have been unaware of the study. In addition, as most participants completed the self-report measures online independently it cannot be confirmed that all participants did have clinical diagnoses of dementia as this was self-reported. Furthermore, although significant differences in self-compassion were not found for gender or dementia subtype, a relatively larger proportion of males and individuals with Alzheimer’s dementia were involved in the study. Also, the measures were only offered in English, even though translated versions of the SCS-SF are available, again limiting generalisability.

Lastly, a variable which would be expected to be uncorrelated with self-compassion (to evidence discriminant validity), such as social desirability (Crowne & Marlowe, 1960), was not included, therefore limiting conclusions regarding overall construct validity of the SCS-SF in dementia. The addition of another measure was decided against at the design and PPI stage as people living with dementia advised that this would create participant burden. Future research should therefore further investigate discriminant validity of the SCS-SF.

Clinical implications and conclusions

On a conceptual basis, the findings highlight the value of self-compassion within clinical work with people with dementia; considering that self-compassion appears to be related to key areas of well-being within dementia, the SCS-SF may be a useful scale to help assess and formulate well-being in clinical settings for people with dementia. However, given the possibility that a two-factor structure of the SCS-SF measures two distinct concepts, it is recommended that when using the SCS-SF with people living with dementia in clinical or research settings, caution is taken when interpreting the total score. Instead, clinicians may wish to interpret the positive and negative factor scores separately. These findings may aid clinicians in considering self-criticism and self-compassion when care planning and supporting people with dementia for example, if future research also suggests that self-compassion is predictive of well-being for people living with dementia, it may be beneficial for interventions aiming to support well-being with people with dementia to cultivate self-compassion but also reduce self-criticism. People with low self-compassion may benefit from self-compassion techniques within Compassion Focussed Therapy and mindfulness, whilst high self-criticism could be addressed using Cognitive Behavioural Therapy techniques such as thought challenging.

To conclude, the SCS-SF shows evidence of reliability and construct validity for people with dementia. However, the scale may comprise positive and negative factors that measure distinct concepts in dementia, and these may be subject to age. Further exploration of the structure and suitability of the SCS-SF to measure self-compassion for people living with dementia is warranted.

Ethical approval

Ethical approval granted by the University of Hull and the Yorkshire and Humber Sheffield Full Research Ethics Committee (IRAS ID 311460).

Disclosure statement

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

Data availability statement

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

Additional information

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.