Compassionate engagement and action scales

The Compassionate Engagement and Action Scales(CEAS; [32]) are based on Gilbert’s evolutionary model of compassion [1,35]. Gilbert [36] conceptualizes compassion as a motivation, defining it as “a sensitivity to suffering in self and others with a commitment to try to alleviate and prevent it” [32, p. 1]. This definition involves two underlying processes presented in a stimulus-response algorithm: first, an awareness or engagement with suffering, and then the motivation or intention to act. Each of these processes (engagement and action) involve distinct skills and competencies. For example, to engage with suffering requires attention sensitivity, empathy, and distress tolerance, while compassionate action requires the skills and wisdom to respond effectively to alleviate suffering [1].

Another defining feature of Gilbert’s model of compassion is understanding the “dynamic reciprocal processing nature of compassion” [37, pp. 26–27], referred to as the flow of compassion [1]. In this model, compassion can be directed from self to other (compassion for others), other to self (compassion from others), and self to self (self-compassion), reflecting inter- and intrapersonal dimensions. Examining the relationship between these different orientations of compassion is particularly relevant in clinical settings, as it has been observed that it is often more difficult for individuals to be compassionate towards themselves than it is to be compassionate towards others [1,38,39].

Given these important defining features of Gilbert’s model, the CEAS consists of three scales, assessing each orientation of the flow of compassion (i.e., compassion for others, compassion from others, and compassion for self). It is proposed that each scale can be used to provide a single-factor score, as well as subscale scores assessing engagement and action [32,40]. The engagement competencies include motivation to care for well-being, attention and/or sensitivity to suffering, sympathy, distress tolerance, empathy, and being accepting and non-judgmental [32]. The action competencies include directing attention to what is helpful, thinking and reasoning about what is likely to be helpful, taking helpful actions, and creating inner feelings of support, kindness, helpfulness and encouragement to deal with distress [32]. Each of the three scales consists of 13 items: six items assessing compassionate engagement, four items assessing compassionate action, and three reverse-formulated items intended to reduce response bias which are excluded from scoring and factor analysis (see S1 Appendix: S1 Table for further details regarding scale development).

Compassion scale

The Compassion Scale (CS; [33,41]) assesses compassion for others, based on Neff’s theoretical model of self-compassion [3,42]. Neff’s conceptualization of self-compassion emphasizes affective and cognitive processes, and involves six components presented as three bipolar continuums: self-kindness and self-judgment; mindfulness and overidentification; common humanity and isolation. Taken together, these six components interact to capture both compassionate (self-kindness, mindfulness, and common humanity) and uncompassionate (self-judgment, overidentification, and isolation) self-responding. This model provides the basis of the most commonly used measure in self-compassion research, the Self-Compassion Scale (SCS; [42]).

While the SCS is widely used, there is debate regarding the inclusion of the negative subscales (capturing uncompassionate self-responding) in a total score. Researchers argue that the positive and negative elements of the scale reflect two distinct factors, which have been shown to have differing effects on quality of life and well-being, and propose using the scores for the positive and negative items separately [12,25,43]. Neuroimaging studies further support the theoretical and statistical arguments that the SCS may be measuring two distinct processes [44], and some state that it is beneficial to study compassionate and uncompassionate self-responding separately to better understand compassion as a protective factor in psychopathology [45,46]. However, Neff supports the use of a total score or six subscale scores rather than separating the positive and negative subscales, arguing that it is common for opposite ends of bipolar continuums to differentially predict outcomes [26,47–49]. There is empirical support for the use of a total score, as well as two scores separating positive and negative items, based on the six-factor theoretical model.

In developing the CS, the authors adapted Neff’s model of self-compassion to be relevant to other-focused attitudes, focusing on emotional responding, cognitive understanding, and attention. The positive elements (kindness, mindfulness, and common humanity) remained similar, while the negative elements were adapted and collapsed into a single factor labelled “indifference.” The final scale has four factors with a total of 16 items (see S1 Appendix: S1 Table for further details regarding scale development).

Sussex-Oxford compassion scales

The Sussex-Oxford Compassion Scales (SOCS; [34]) assess compassion for self (SOCS-S) and compassion for others (SOCS-O), using a five-factor structure based on a theoretical model proposed by Strauss et al. [24]. In response to the lack of consensus on a definition of compassion, Strauss and colleagues conducted a qualitative review in an effort to consolidate a range of different conceptualizations and generate an operational definition that includes cognitive, affective, and behavioural dimensions. Based on the qualitative review, the authors propose a definition of compassion consisting of five elements: recognizing suffering; understanding the universality of human suffering; feeling for the person suffering; tolerating uncomfortable feelings; and motivation to act/acting to alleviate suffering. They then conducted a systematic review of self- and observer-rated measures of compassion and rated the extent to which measures assessed each of the identified five elements. This five-element model was later empirically tested with a series of factor analytic studies with existing compassion measures, which supported the five-factor structure, however also identified limitations with existing measures and identified a need for the development of new measures [50]—thus, the SOCS were developed.

Items were generated through interviewing experts in contemplative approaches (N = 22, 72.7% women)—researchers defined experts as researchers or teachers in the fields of mindfulness and compassion, and included at least one researcher and one teacher from each of the six continents. Experts generated 155 items assessing compassion for others, and 101 items assessing self-compassion, which were then reduced to 120 total items (60 other, 60 self) through iterative review. Items were subsequently reviewed by 15 experts and 15 nonexperts (undergraduate students, 60% women), who were asked whether the items represented the element for which they were intended. Item reduction resulted in 57 other-compassion and 58 self-compassion items. The next stage involved further item reduction using CFA with 1,017 health care workers (79.6% women), in which the four highest loading items for each of the five proposed factors were retained, resulting in 20 items assessing compassion for others, and 20 items assessing self-compassion (see S1 Appendix: S1 Table for further details regarding scale development).

Compassion and other psychological constructs

The relationships between compassion and other psychological constructs inform a key aspect of the development of each compassion measure. In assessing convergent validity, authors for each scale selected measures based on literature demonstrating relationships between compassion (for self, for others, and from others) and constructs related to self-/other-relating as well as mood and well-being. One key observation is that the relationship between compassion for self (or self-compassion) and compassion for others is unclear, and often studies examining these constructs have found only weak or insignificant correlations [13,51]. This finding was consistent for the studies developing the CEAS and CS, whereas results from the SOCS found that the SOCS-S and SOCS-O showed medium associations, which emphasizes the importance of advancing our understanding of the relationship between the different orientations of compassion. The Self-Compassion Scale [42] was included in all three studies. Gilbert et al. [32] used two scores, separating positive and negative items, while Pommier et al. [33] and Gu et al. [34] used a total score.

The CEAS and SOCS included the DASS-21 [52,53], assessing depression, anxiety, and stress, and the Warwick-Edinburgh Mental Well-Being Scale [54], assessing well-being. The relationship between self-compassion and symptoms of psychopathology is well documented, suggesting that self-compassion is associated with lower levels of psychopathology [20,55] and enhanced well-being [17,56,57]. The relationship between compassion for others and these constructs is less clear, and thus examining correlations with new measures of compassion for others presents an opportunity to advance understanding in this area [13,58]. The authors of the SOCS included the Five Facet Mindfulness Questionnaire [59], excluding the “observing” items. Mindfulness and compassion are commonly discussed in tandem, and while authors emphasize the importance of distinguishing between these different processes [2,60,61], mindfulness is a proposed factor in the SCS and CS. Thus, we would anticipate that measures of compassion based on models that explicitly include mindfulness would be positively associated with a measure assessing this separate but overlapping construct. The authors of the CS included a measure of attachment to assess discriminant validity, as attachment represents a way of measuring a self-other schema that is related but distinct from compassion for others [33]. Similarly, self-compassion has been demonstrated to be related to attachment style [18,62–65], which can be theoretically understood from an evolutionary perspective—individuals develop the capacity for caring motives through early experiences with caregivers, influencing neurobiology and subsequent ways of relating to self and others.

Objectives

The objectives of the current study are to translate each of these measures from the original English to French, analyze the factor structure proposed by each respective theoretical model, and examine the relationship between compassion (to self, to others, and from others) and other related constructs. The ultimate goal is to expand the use of these measures to French-speaking populations, thereby contributing to a more complete understanding of compassion that accounts for multicultural and linguistic diversity. The importance of having valid and reliable measures in more than one language is key to advancing psychological research beyond English-speaking contexts, and is especially important when conducting studies and ensuring accessibility in multilingual contexts.

Translation

The English versions of the CEAS, CS and SOCS were translated to French using a five-stage process [66–68], with permission from the authors of the original English scales. In stage one (initial translation), items of all scales were forward-translated (English to French) independently by two bilingual native-French speakers (one translator familiar with the scale content, and one who was naïve to the scale objective). During stage two (synthesis), discrepancies between translations were discussed between the two translators, and resolved with the assistance of a third native-French-speaking research assistant. In stage three (back translation), the synthesized translations were translated from French back to English by two bilingual native-English speakers. In stage four (expert committee), a committee consisting of the principal investigator, co-investigators, and translators met to consolidate all versions of the scale and produce a pre-final version. Decisions were documented in translation reports, which were shared along with the pre-final version of the translated scale with authors of the English scales for feedback and to ensure scale equivalency (see S2 Appendix: Translation Reports). Time constraints prevented such feedback regarding the pre-final French version of the CS.

During the final stage (testing), the pre-final scales were tested with a small sample of native-French speakers (n = 7) to provide feedback regarding the test-takers’ experience and understanding of the scale items. Test-takers were asked to complete all translated scales online via the survey platform, Qualtrics. They had not previously reviewed the English scales. Upon completion, they were offered the option to provide feedback either verbally via a phone conversation with the PI, or in written form via email. All individuals chose to speak with the PI. Questions/prompts included, “please describe your experience completing the questionnaires,” “was there anything that was unclear,” and “do you have any comments on the phrasing of the items?” Overall, feedback was positive (i.e., the questionnaires were clear and easy to understand) without any proposed changes to the phrasing of the items. However, participants shared that they experienced reading the introductory sections for the CEAS scales as repetitive, as background information with details defining compassionate engagement and action is repeated for each of the three scales. Based on this feedback, and with permission from the authors of the English CEAS, the full directions for the CEAS appeared only in the first scale (CEAS-SC), and were shortened for the subsequent scales (CEAS-TO and CEAS-FROM; see S2 Appendix: Translation Reports for details regarding these edits).

Participants and recruitment

Ethics approval was obtained from McGill University’s Research Ethics Board (REB File Number: 19-11-034). Participants were recruited via online advertising distributed via university listservs and social media platforms. Inclusion criteria included identifying French as one’s primary or native language and being 18 years or older. Participants provided informed written consent prior to completing an online survey using the platform Qualtrics. Recruitment and data collection took place from January 2 to July 29, 2021. Determining adequate sample size was informed by recommendations that a minimum sample should consist of 10 observations per scale item [69–72], as well as guidance that most models with fewer than 40 items can be tested with a minimum sample of 200 participants [73–75]. Comrey and Lee [76] provide classifications based on absolute sample size, rating a sample of 100 as poor, 200 as fair, 300 as good, 500 as very good, and 1,000 as excellent. Given that the longest scale in the present study consists of 20 items (SOCS-S/SOCS-O), a minimum sample of 200 participants is required.

Compassionate Engagement and Action Scales (CEAS)

The CEAS [32] consist of three scales, each with 13 items assessing (1) self-compassion (CEAS-SC), (2) compassion to others (CEAS-TO), and (3) compassion from others (CEAS-FROM), i.e., the ability to receive compassion from important people in an individual’s life. On each scale, the individual is asked to rate how a statement applies to themselves when they or the people in their lives are distressed. Responses are reported using a scale of 1 (never) to 10 (always).

Analyses

All statistical analyses were performed on R 4.2.1. The analysis script and dataset are available via the Zenodo Repository [88,89].

Demographics

See Table 1 for complete demographic information. Participants ranged in age from 18 to 72 years (median age = 30 years), and the majority identified as women. In terms of ethnicity, participants were asked to enter a text response, and the majority identified as White. Most participants reported learning French in Quebec, followed by France.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic information.

https://doi.org/10.1371/journal.pone.0305776.t001

Confirmatory factor analysis

Compassionate engagement and action scales.

Fit indices for all models are shown in Table 2. Path diagrams are available in supporting information (S1-S9 Figs in S2 Appendix). Fit indices for one-factor and two-factor models for the CEAS-SC, assessing self-compassion, showed poor fit. A three-factor, non-hierarchical model showed acceptable fit across fit indices, assessing sensitivity to suffering, engagement with suffering, and compassionate action. The three-factor, three-level hierarchical model proposed by the scale authors, in which sensitivity to suffering and engagement with suffering load onto “compassionate engagement,” did not converge on a solution. As such, the three-factor, non-hierarchical model can be interpreted as best fitting the data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Model fit indices for the compassionate engagement and action scales.

https://doi.org/10.1371/journal.pone.0305776.t002

For compassion to others (CEAS-TO), fit indices for the one-factor model suggest poor fit. The two-factor, non-hierarchical model shows acceptable fit based on the NNFI, CFI, and SRMR, while RMSEA suggests poor fit. The scale authors proposed a two-factor hierarchical model, however this model’s variance-covariance matrix was not positive definite, suggesting that the model is not identified. The two-factor, non-hierarchical model can be interpreted as best fitting the data, assessing compassionate engagement and compassionate action.

For compassion from others (CEAS-FROM), fit indices for the one-factor model show acceptable fit, with the exception of RMSEA. Fit indices for the two-factor model show acceptable fit, as do fit indices for the two-factor, hierarchical model.

Compassion scale.

Table 3 shows fit indices for all models. Path diagrams are available in supporting information (S10-S12 Figs in S2 Appendix). All models (one-factor, four-factor, and four-factor hierarchical) show acceptable fit across fit indices. The four-factor and four-factor hierarchical models both show acceptable fit using conservative cutoffs. RMSEA and SRMR values for the four-factor model appear slightly lower, suggesting a four-factor model best fits the data; however, differences are small. The four-factor hierarchical model proposed by the scale authors is supported by fit indices in the present sample.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Fit indices for the compassion scale.

https://doi.org/10.1371/journal.pone.0305776.t003

Sussex-Oxford compassion scales.

Table 4 shows fit indices for all models. Path diagrams are provided in supporting information (S13-S18 Figs in S2 Appendix). For the Sussex Oxford Compassion Scale for Self (SOCS-S), fit indices for a one-factor model suggest poor fit. A five-factor, non-hierarchical model is supported by CFI, RMSEA, and SRMR, which all suggest acceptable fit, while NNFI suggests poor fit. Similarly, a five-factor hierarchical model appears acceptable based on CFI, RMSEA, and SRMR, with NNFI failing to meet the cutoff criteria. Both the five-factor and five-factor hierarchical models are supported in this sample.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Fit Indices for the Sussex-Oxford compassion scales.

https://doi.org/10.1371/journal.pone.0305776.t004

For the Sussex-Oxford Compassion Scale for Others (SOCS-O), fit indices for the one-factor model all suggest poor fit. The five-factor non-hierarchical model failed to converge on a solution. The five-factor hierarchical model appeared acceptable across fit indices, successfully replicating the model proposed by the scale authors.

Internal consistency

See Table 5 for descriptive statistics, Cronbach’s alpha, and omega total estimate. Omega total estimates were calculated using standardized item loadings from two-factor hierarchical models for CEAS-TO and CEAS-FROM, and a three-factor hierarchical model for CEAS-SC. Omega total estimates for all subscales and total scores exceeded the 0.70 cutoff other than “sensitivity to suffering,” a sub-factor of compassionate engagement in the CEAS-SC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Descriptive statistics and internal consistency.

https://doi.org/10.1371/journal.pone.0305776.t005

For the CS, omega total estimates were calculated using standardized item loadings from the four-factor hierarchical model, and showed overall good internal consistency. Cronbach’s alpha ranged from 0.64 to 0.85. The only subscale that did not exceed the 0.70 cutoff for both Cronbach’s alpha and omega total estimate was the “Indifference” (reverse-scored) subscale.

For both SOCS-S and SOCS-O, omega total estimates were calculated using standardized item loadings from five-factor hierarchical models. Omega total estimates showed good internal consistency. The only score below the 0.70 cutoff was “Tolerating uncomfortable feelings,” a subscale of the SOCS-O.

There were no floor or ceiling effects (see S1 Appendix: S2 Table).

Measurement invariance

Table 6 shows results from multi-group analyses to assess measurement invariance. Results for the CEAS-SC support measurement invariance for the three-factor model: CFI and RMSEA met the cut-off criteria. Results for CEAS-TO and CEAS-FROM supported configural, metric, and scalar invariance for the two-factor hierarchical models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Measurement invariance analyses based on age group.

https://doi.org/10.1371/journal.pone.0305776.t006

Analyses for measurement invariance for the CS for the four-factor hierarchical model indicated potential model misspecification, and results could not be reported. Invariance analyses for the four-factor non-hierarchical model showed good fit indices and comparison indices for CFI, while RMSEA failed to meet the cut-off. As such, the CS is not considered invariant across age groups.

Similarly, invariance analyses for the SOCS-S and SOCS-O indicated possible issues of misspecification, and these scales are not considered invariant across age groups.

Chi-square (χ²) tests are reported for descriptive purposes, but were not interpreted due to sensitivity to sample size. Invariance analyses using alternate demographic variables (e.g., gender, education level) were not possible due to insufficient sample size.

Convergent validity

Compassionate engagement and action scales.

Correlation coefficients between total scale scores are shown in Table 7 (see S1 Appendix: S3 Table for an expanded correlation matrix including subscale scores). The CEAS-SC was found to have a moderate positive correlation with compassion from others (CEAS-FROM), and large positive correlations with other measures assessing self-compassion (SOCS-S, SCS-POS), mindfulness (FFMQ), and mental well-being (WEMWBS). Positive correlations with a small effect size were found between the CEAS-SC and measures assessing compassion for others (CEAS-TO, CS, SOCS-O). Moderate, negative correlations were found with attachment insecurity (ECR-Anx, ECR-Avoid), and large negative correlation with psychological distress (DASS-21) and uncompassionate self-responding (SCS-NEG).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Pearson correlation coefficients adjusted by Benjamini-Hochberg procedure.

https://doi.org/10.1371/journal.pone.0305776.t007

The CEAS-TO was found to have large, positive correlations with other measures assessing compassion to others (CS, SOCS-O), and a small, positive correlation with compassion from others (CEAS-FROM), self-compassion (SOCS-S), mindfulness (FFMQ), and mental well-being (WEMWBS). There was a small negative association with attachment avoidance (ECR-Avoid), but no significant correlation with attachment anxiety (ECR-Anx). The CEAS-TO did not significantly correlate with the positive or negative composite scores of the Self-Compassion Scale (SCS-POS, SCS-NEG), and did not significantly correlate with measures assessing psychological distress (DASS-21).

The CEAS-FROM was found to significantly, positively correlate with a moderate effect size with measures assessing self-compassion (CEAS-SC, SOCS-S) and mental well-being (WEMWBS). Small, significant positive correlations were found with a third measure of self-compassion (SCS-POS), as well as with compassion to others (CS, SOCS-O) and mindfulness (FFMQ). Moderate, negative correlations were found between CEAS-FROM and measures assessing uncompassionate self-responding (SCS-NEG), psychological distress (DASS-21), and attachment avoidance (ECR-Avoid), and a small negative correlation with attachment anxiety (ECR-Anx).

Factor structure

Measurement invariance

We assessed measurement invariance for all scales across age (median split). Results showed support for measurement invariance for the CEAS-SC, CEAS-TO, and CEAS-FROM, and partial support for the CS. We were not able to assess measurement invariance for the SOCS-S or SOCS-O due to possible misspecification. Measurement invariance assesses whether a questionnaire has the same structure and meaning across groups, indicating whether scores can be compared across groups (in this case, age). Sample size posed a limitation to analyzing measurement invariance in the present study, contributing to the possible misspecification issues for the SOCS-S/O and four-factor hierarchical model of the CS. A larger sample and/or additional datasets would enable further analyses to assess whether the issues in the present study are due to sample size, or whether there are significant differences in how different demographic groups respond to the questionnaires. Future research should examine measurement invariance across additional demographic groups, including gender and education level.

Convergent validity

Convergent validity analyses provide insight into the different orientations of compassion and their association with other constructs. Correlations between measures assessing self-compassion and other measures were consistent with predictions, and in line with the broader literature. The SOCS-S total score and the CEAS-SC were significantly correlated in the predicted direction with all other measures. These findings are consistent with existing literature supporting the association of self-compassion with higher levels of mindfulness and well-being, and lower levels of psychological distress and insecure attachment [17,18,24,116–119].

The total score for the CEAS-FROM was significantly correlated with all measures in the predicted direction. These findings suggest that individuals who are more able to receive compassion from others experience higher levels of self-compassion, compassion for others, mindfulness, and well-being, and lower levels of psychological distress, attachment insecurity, and uncompassionate self-responding. Correlations with measures of compassion for others were small, which suggests that the ability to receive compassion from others may not be strongly related to the ability to show compassion to other people. Theoretically, this observation fits within the context of research demonstrating that individuals with insecure attachment styles may feel more comfortable showing care for others than receiving care [120,121], and that these competencies may be distinct. Results can also be understood in the context of research on fears of compassion, in which fear of receiving compassion has a stronger association with mental health outcomes than fears of being compassionate to others [37]. These findings highlight the importance of further research on the flow of compassion and understanding the relationship between different orientations of compassion.

The relationship between compassion for others and other orientations of compassion is in line with our predictions, and consistent with existing literature. Generally, all measures of compassion for others (CEAS-TO, CS, SOCS-O) were associated with one another as well as with at least two of the measures assessing self-compassion. The measures assessing compassion for others were not significantly correlated with measures assessing uncompassionate self-responding (SCS-NEG), psychological distress, nor attachment anxiety, but showed small negative correlations with attachment avoidance. Findings related to the relationship between compassion for others and mental health are markedly different from findings related to self-compassion and psychopathology, and in line with previous research [13]. The lack of relationship between compassion for others and uncompassionate self-responding (SCS-NEG) is in line with the common observation that individuals are more likely to be compassionate to others than to themselves, and could be harsh towards themselves irrespective of how they respond to others. However, the CS was not significantly correlated with SCS-NEG, which is somewhat surprising given that the originally proposed three negative subscales (later collapsed into one subscale labelled “indifference”) were adapted from the three negative subscales of the SCS. These findings could be understood in the context of social desirability bias, as when asked how one would respond to another’s suffering on a questionnaire, individuals may be more likely to report that they would be compassionate toward others, whereas they may be more honest about being uncompassionate to themselves on the self-compassion measures.

The relationship between compassion to others and attachment insecurity is somewhat surprising, as the way that an individual is cared for and the types of caring relationships that are modelled to them by caregivers seem like important factors associated with how individuals relate to others. While our results support a theoretical relationship with attachment avoidance, but not attachment anxiety, it is noteworthy that the subscales of the ECR-R are not considered distinct orthogonal constructs—they are conceptually overlapping [122], and thus these results should not be over-interpreted. The differing relationship between attachment anxiety and avoidance may be due to the phrasing of the scale items, as all the items in the ECR-Anxiety subscale are phrased negatively, while the majority of the items in the Avoidance subscale are phrased positively and reverse-scored. It is particularly interesting that self-compassion appears to have a more robust relationship with attachment insecurity, highlighting the importance of caregiving relationships in developing a compassionate style of relating to oneself.

Implications and limitations

Findings from the present study support the use of the translated measures using the factor structures proposed by the original English studies, with the exception of the CEAS-SC and CEAS-TO, for which CFA findings for two-factor models were supported. The two-factor non-hierarchical model is still theoretically aligned with the conceptualization of “engagement” and “action” reflecting distinct processes. Internal consistency and convergent validity results support the use of a total score for all measures.

Each of the measures reflects a distinct theoretical conceptualization of compassion, and furthermore, the measures use different approaches to developing these theoretical conceptualizations. Gilbert [123] distinguishes between a stimulus-response algorithm that identifies distinct skills and competencies in two different stages (engagement and action), versus a cluster-diagnostic approach which identifies core processes that are commonly understood as related but may vary in how they group together. He identifies this as a potential limiting factor of relying on factor analysis to define a given phenomenon. The goal of the present study was to provide valid measures for use with French populations to further study compassion using these different models, and not to identify a “correct” model. The model chosen necessarily depends on the theory underpinning a given study [27]. Recent studies on compassion using alternate forms of measurement, for example physiological measures such as heart-rate variability, provide important additional data to this debate [124–129].

There are several limitations to the present study. While a thorough forward- and back-translation process was undertaken including consultation with authors of the original source scales, the pilot stage was conducted with a small group of participants and did not include structured focus groups and/or cognitive interviews. Incorporating a more thorough qualitative stage into the translation process could have strengthened the semantic equivalence of the translated scales [130]. Additionally, while our sample size surpasses the minimum criteria for factor analysis [76], issues with model fit for the CEAS-SC and CEAS-TO may be impacted by sample size due to the relative complexity of the proposed models. Sample size also negatively impacted measurement invariance analyses, limiting the demographic variables that could be used and potentially contributing to issues of misspecification for the CS and SOCS. Future studies examining measurement invariance for these scales should consider a larger sample size as well as including a comparative sample using the original scale language and culture [131]. These added steps would provide further insight into the equivalence, meaning, and interpretation of the scales in diverse cultural and linguistic contexts.

Finally, representativeness and generalizability should be considered. While participants in the present study were a convenience sample and not a representative sample of the Quebec population, representation based on ethnicity, age, and level of education are comparable to the most recent census data [132]. The average age of the population in Quebec is 42.80 years—our sample is slightly younger, at 34.97 years. Comparing ethnicity in our sample to the Quebec population, 91.05% of our sample self-identified as White compared to 83.86% of the Quebec population identifying as “not a visible minority.” Level of education attainment in the current sample also appears comparable to the Quebec population: 36.87% of the Quebec population reported holding a postsecondary diploma/certificate/degree below bachelor’s as their highest level of education, and 43.62% of our sample reported the same, while 23.54% of census participants reported holding a bachelor’s degree compared to 29.87% in our sample. The present sample predominantly identified as women (86.79%), which is not representative of the Quebec population, but is consistent with samples used for the development of the English scales. The samples involved in the development and validation of the original scales included 62% to 88% females, and 50% to 85% of participants identified as Caucasian [33] or were described as “predominantly White” in study reports where numerical information was not reported [34]. Of the 13 samples used to assess factor structure, seven samples comprised students, two comprised health care workers, three were community samples (of which two were recruited from Mechanical Turk), and one comprised meditators (see S1 Appendix: S1 Table).

Regarding generalizability, while French speakers in Canada reflect a heterogenous and culturally diverse population, there may be cultural and linguistic differences with French-speaking populations in different geographic areas. Future studies validating the use of these translations in other regions would be valuable. Lastly, participants completed an online questionnaire with a significant number of items assessing similar constructs, which may have generated response biases—for example, the tendency to be consistent.