Exploring compassionate attributes and skills among individuals participating in compassion-focused therapy for enhancing well-being.

OBJECTIVES: The conceptual approach of compassion underlying compassion-focused therapy (CFT) is based on theoretical rather than empirical grounds. The aim of the present study was to seek empirical support for components of compassion as outlined in the theoretical model underpinning CFT, and to explore which components, if any, matter most for improving well-being.
DESIGN: A sequential exploratory mixed methods design was employed.
METHODS: Alongside a randomized controlled trial (RCT), we systematically examined 625 emails sent by 87 RCT participants to five counsellors during the course of a well-being enhancing CFT self-help intervention, to identify theoretically based compassionate attributes and skills. Next, in a quantitative analysis, we compared participants who did and did not show clinically relevant improvement on well-being with regard to the occurrence of compassionate attributes and skills.
RESULTS: Although the theoretical model of compassion integral to CFT was largely supported by the emails, it was slightly simplified so as to better fit the data. The adjusted model comprises five compassionate attributes (i.e., care for well-being, sensitivity, empathy, distress tolerance, and common humanity) and four compassionate skills (i.e., compassionate attention, reasoning, behaviour, and feeling/sensation). Three illustrative cases are presented to contribute to a better understanding of fundamental components of compassion. Quantitative analyses indicate that participants showing clinically relevant improvement on well-being expressed significantly more compassionate feeling/sensation compared to those who did not.
CONCLUSIONS: We found preliminary evidence for the conceptualization of compassion underlying CFT. Compassionate feeling/sensation bears particular interest when well-being is the intended outcome of CFT. PRACTITIONER POINTS: Individuals participating in compassion-focused therapy for enhancing well-being experience a wide range of compassionate attributes and skills. Compassion-focused therapy may instigate well-being if a client is able to experience compassionate feeling/sensation.

RCT Entities:   Population Interventions Outcomes

Background

Over recent years, increasing attention has been drawn to compassion as a key resource for mental health. In brief, compassion can be understood in terms of the ability to be sensitive to the suffering of self and others combined with a commitment to try to alleviate or prevent it (Gilbert, 2014b). The growing interest for compassion gave rise to the development and evaluation of various compassion‐based interventions of which compassion‐focused therapy (CFT; Gilbert, 2009, 2014b) currently is the best evaluated (Kirby, 2016; Kirby, Tellegen, & Steindl, 2017; Leaviss & Uttley, 2015). CFT has been shown to elicit favourable effects on psychological symptoms and well‐being in multiple studies (e.g., Braehler et al., 2013; Gilbert & Procter, 2006; Shapira & Mongrain, 2010; Sommers‐Spijkerman, Trompetter, Schreurs, & Bohlmeijer, 2018aa).

Although the cultivation of compassion has been postulated as the primary working mechanism behind the positive effects of CFT on mental health, there is as yet little empirical evidence to support this. Even more limited is research on change processes which account for improvements in compassion during CFT. Therapy change process research is deemed a valuable complement to effectiveness research in randomized controlled trials as it may shed further light on how and why CFT works in some individuals and not in others (Elliott, 2010). To aid advances in therapy change process research in this emerging field, it is essential to increase empirical knowledge of key processes of compassion.

In his dual theoretical model of compassion, Gilbert (2009, 2014b) outlines various underlying processes of compassion. The model posits that, to develop a compassionate self, one must adopt two mindsets or ‘psychologies’, encompassing a multitude of compassionate attributes and skills. The first mindset involves the motivation and ability to notice, engage with, and make sense of the suffering of self and others. Six engagement attributes are distinguished pertaining to this first dimension of compassion (i.e., sensitivity to suffering), that is, (1) care for well‐being, the motivation/willingness to address suffering and/or facilitate flourishing; (2) sensitivity, the ability to recognize suffering; (3) sympathy, the capacity to feel emotionally connected to suffering; (4) distress tolerance, the ability to stay with and tolerate rather than avoid or deny emotions evoked by suffering; (5) empathy, the ability to stand back from and understand suffering; and (6) non‐judgement, the ability to take an accepting, non‐critical, and non‐condemning approach. According to Gilbert's model, the second mindset of compassion involves the skills and wisdom to undertake actions towards preventing or alleviating suffering of the self and others. This requires a set of six transformative skills, including (1) compassionate attention, the ability to pay attention to what is helpful and supports us; (2) compassionate reasoning, the ability to use supportive, reassuring thought patterns; (3) compassionate behaviour, the ability to act upon suffering in a way that alleviates distress and facilitates development and growth; (4) compassionate imagery, the ability to apply imagery and meditation‐like practices in order to cultivate affiliative emotions; (5) compassionate feeling, the ability to experience emotions linked to compassion; and (6) compassionate sensation, the ability to generate physical states conducive to compassion.

Although several studies indicate that CFT may contribute to greater levels of compassion (Arimitsu, 2016; Matos et al., 2017), little empirical knowledge is available on these underlying attributes and skills. In recent years, a handful of studies have explored fundamental processes underlying the inner transformation from a self‐critical towards a self‐compassionate self, whether or not in relation to CFT (Lawrence & Lee, 2014; Waite, Knight, & Lee, 2015). Using interpretative phenomenological analysis of semi‐structured interviews, these studies disclose a number of processes which foster a compassionate mind. In seven patients with post‐traumatic stress disorder (PTSD) who had completed a CFT course for trauma, Lawrence and Lee (2014) identified five superordinate themes with regard to the development of a self‐compassionate mind, including the battle to give up the inner critic, the aversive and alien experience of developing compassion; the emotional experience of therapy; self‐compassion as a positive emotional experience, and a more positive outlook in the present and for the future. In another study, Waite et al. (2015) revealed processes of self‐compassion in patients in recovery from psychosis. Compassionate processes that came to light were among others acceptance of oneself and one's illness, treating oneself with kindness, and finding hope for the future. Although these studies make the process of becoming self‐compassionate more insightful, the observations made are based on very small samples and focus on rather specific, clinical populations. Moreover, despite that some of the reported processes touch upon attributes and skills of compassion as defined by Gilbert (2009, 2014a, 2014b) such as distress tolerance and non‐judgement, these studies do not comprehensively address all aspects of the theoretical model integral to CFT.

The present study sought to provide empirical support for the entire theoretical model of compassionate attributes and skills in a more generic population. In order to do so, the study built upon a two‐arm randomized controlled trial (RCT) investigating the effectiveness of CFT as guided self‐help in a self‐selected Dutch community sample with low to moderate levels of well‐being (Sommers‐Spijkerman et al., 2018aa). Findings from the RCT demonstrated that despite overall effectiveness of the intervention at group level, individual participants responded differently to the intervention. Self‐compassion emerged as a mediator of the effects of CFT on well‐being (Sommers‐Spijkerman, Trompetter, Schreurs, & Bohlmeijer, 2018bb). However, these findings remain inconclusive as to which compassionate processes precisely bring about improvements in well‐being.

The intervention group of the RCT followed a 9‐week CFT‐based self‐help intervention for enhancing well‐being. During the intervention, they had weekly email contact with a counsellor to discuss their experiences with the book and address any questions or barriers. Through systematically analysing emails sent by participants to their counsellor, we aimed to examine to what extent compassionate attributes and skills underlying the two mindsets of compassion integral to CFT are empirically distinguishable. To provide a meaningful lens from which to understand the compassionate attributes and skills, a number of illustrative cases from the trial are presented. A secondary aim was to explore to what extent it is possible to discriminate between participants who improved on well‐being over the course of the intervention and those who did not based on frequencies of compassionate attributes and skills.

Materials and methods

Design

We used data collected in the context of a recent waitlist RCT investigating the effects of CFT as guided self‐help on well‐being in a Dutch community sample (Sommers‐Spijkerman et al., 2018aa). CFT was offered as guided self‐help including email counselling. The present study employed a sequential exploratory mixed methods design (Creswell & Plano Clark, 2007; Hanson, Creswell, Clark, Petska, & Creswell, 2005). In the first step, email messages sent from the participants in the CFT group to their counsellor throughout the intervention period were content‐analysed to identify compassionate attributes and skills as defined by Gilbert (2009, 2014b). A qualitative exploration of a number of individual cases served to further deepen the reader's understanding of attributes and skills underlying compassion revealed by the content analysis. This is followed by the quantitative part of the study, in which we compared participants who did and did not show clinically relevant improvement on well‐being with regard to the occurrence of the different attributes and skills.

This study was approved by the Faculty of Behavioural Sciences Ethics Committee at the University of Twente (BCE17382). Informed consent was obtained from all participants, also regarding the use of sent emails for scientific purposes.

Participants

The RCT sample comprised 242 participants with low to moderate levels of well‐being, as determined by the Mental Health Continuum‐Short Form (MHC‐SF; Keyes, 2002; Lamers, Westerhof, Bohlmeijer, ten Klooster, & Keyes, 2011), of whom 120 were allocated to CFT and 122 to a waitlist control condition. For the current study, we used data from all participants in the CFT group who completed the CFT self‐help intervention and reflected on each lesson in the email correspondence with the counsellor within a 9‐week intervention period. Completion of the self‐help intervention was determined through one question in an intervention evaluation questionnaire which was administered among the CFT group at post‐measurement (i.e., 3 months after baseline). Participants were asked to indicate per lesson whether or not they completed the lesson. To determine whether participants who completed the entire intervention also reflected on each lesson in their email correspondence with the counsellor, we screened the emails. The majority of the sample (n = 87) was female (80.5%), high‐educated (81.6%), and in paid employment (78.2%) (Table 1). Mean age of the participants was 51.92 years (SD = 9.63).

Table 1

Demographic characteristics of the participants (n = 87)

Socio‐demographics	Total (n = 87)	Improversa (n = 42)	Non‐improversb (n = 45)	p c
Age, years
M (SD)	51.92 (9.63)	51.67 (10.58)	52.16 (8.77)	.815
Range	20–69	20–69	33–68
Gender, n (%)
Male	17 (19.5)	10 (23.8)	7 (15.6)	.332
Female	70 (80.5)	32 (76.2)	38 (84.4)
Nationality, n (%)
Dutch	87 (100.0)	42 (100.0)	45 (100.0)	–
Other	–	–	–
Marital status, n (%)
Married/registered partnership	45 (51.7)	22 (52.4)	23 (51.1)	.906
Not married (never married, divorced, widowed)	42 (48.3)	20 (47.6)	22 (48.9)
Living situation, n (%)
With partner	56 (64.4)	25 (59.5)	31 (68.9)	.362
Without partner	31 (35.6)	17 (40.5)	14 (31.1)
Education level (highest level completed), n (%)
Low (primary school, lower vocational education)	–	–	–	.879
Intermediate (secondary school, vocational education)	16 (18.4)	8 (19.0)	8 (17.8)
High (higher vocational education, university)	71 (81.6)	34 (81.0)	37 (82.2)
Work situation, n (%)
Paid employment	68 (78.2)	34 (81.0)	34 (75.6)	.540
No paid employment	16 (18.4)	6 (14.3)	10 (22.2)
Student	3 (3.4)	2 (4.8)	1 (2.2)

Improvers are participants who showed a clinically relevant improvement on well‐being between baseline and post‐intervention.

Non‐improvers are participants who did not show a clinically relevant improvement on well‐being between baseline and post‐intervention.

p‐values of chi‐square tests and independent t‐tests comparing improvers and non‐improvers.

The intervention

The CFT intervention consisted of the self‐help book ‘Compassion as key to happiness’ (Hulsbergen & Bohlmeijer, 2015) and weekly email counselling. The book comprises seven lessons based on the CFT principles (Gilbert, 2014b) that had to be completed in 9 weeks. Each lesson consists of psycho‐educational information on a key component of compassion and a broad variety of self‐reflective and experiential exercises to cultivate compassionate attributes and skills. Participants were instructed to complete one lesson per week in sequential order. They received email guidance from a counsellor on a weekly basis. All participants were randomly assigned to one of five counsellors. Counsellors were the first and second author, one graduated psychologist and two Master students Psychology. All five counsellors were trained and supervised by two experienced health care psychologists (i.e., fourth and fifth author). During their training, the counsellors studied the self‐help book, performed the exercises, and practised writing emails in the roles of both participant and counsellor. Counsellors and supervisors met once a week for supervision, during the intervention phase, to discuss specific topics and cases brought in by the counsellors. Participants were asked to introduce themselves in the first email and, more specifically, to tell something about their background and motivation to engage in the self‐help intervention. Subsequently, participants were requested to send an email about their progress and experiences once a week, generally after completing a lesson. Participants received a response from their counsellor on a fixed day of the week. The counsellors were instructed to positively reinforce participants, to provide feedback, to answer questions, to prompt participants who did not send an email, and to introduce the theme of the next lesson. A more extensive description of the intervention can be found in Sommers‐Spijkerman et al., (2018aa).

Data analyses

The data consisted of 625 email messages sent by 87 CFT participants to their counsellors during the 9‐week intervention period. Participants sent on average 7.18 email messages to their counsellor (SD = 0.67, range: 5–9). The emails were analysed using both qualitative and quantitative techniques.

Content analysis

Directed qualitative content analysis (Hsieh & Shannon, 2005) was used with the aim to identify attributes and skills underlying compassion reflected in the participants’ email messages and ultimately find evidence for the theoretical model underlying CFT. Participants’ emails were systematically organized using a coding scheme. This coding scheme was developed using a deductive approach; that is, we used pre‐defined compassionate attributes and skills derived from the theoretical model outlined in Gilbert (2014b). For relevant data that could not be coded using these predetermined categories, it was determined whether they represent a new category. Prior to the coding, all email correspondence was anonymously saved in word documents labelled with a unique respondent number. All information that is traceable to the participant was removed from the email conversations, for example participants’ own or family‐member names and places of residence. The first author created a draft coding scheme. This draft coding scheme was tested by the first and second author, independently from one another, on 58 emails written by eight random participants, whereby each attribute and skill was counted once per email message (1 = present, 0 = absent), with a maximum count of 9.

In this and subsequent coding rounds, respondent numbers were randomly drawn by the third or fourth author. Both coders were blind to the identity of the counsellor as well as to intervention outcomes. After the first coding round, the scheme was adjusted and tested again by both coders on the emails (n = 42) of six other participants. In a similar third coding round, consensus was reached regarding the coding scheme and inter‐rater reliability was deemed satisfactory (Cohen's kappa = .68; Landis & Koch, 1977). The final coding scheme (see Tables 2 and 3) was applied by the first author to code all email messages written by the participants to their counsellor. For each participant, messages were analysed in chronological order. Two emails sent briefly after one another, without a response from the counsellor in between, were counted as one message. Email messages concerning illness or delay, in which the participant did not relate to the contents of the intervention, were not analysed. When there was doubt about the assignment of a particular code, the second author was consulted.

Table 2

Compassionate attributes and skills

Compassionate attribute/skill	Definition	%
		Total	I	NI
Care for well‐being	Motivation/intention to be caring for the purpose of alleviating suffering and facilitating flourishing of the self or others	36.3	36.2	36.4
Sensitivity	Ability to be attentive to and recognize (stimuli that trigger) suffering of the self or others (e.g., unpleasant feelings)	53.1	49.2	56.8
Distress tolerance	Ability to contain, stay with, and tolerate suffering/distress (e.g., unpleasant feelings, memories or situations) rather than avoid, fearfully divert from, close down, contradict, invalidate, or deny them	7.8	10.0	5.9
Empathy	Ability to stand back from and view our own or others’ suffering and its causes from the perspective of someone else or different parts of ourselves (cognitive empathy) and/or to emotionally connect with the suffering of self or others (emotional empathy/sympathy). This concerns a reflection on/thoughts and feelings underlying suffering/distress and does not involve a shift in thinking styles	46.2	45.5	46.9
Common humanity	Ability to recognize that suffering is part of the human experience and that we are not alone in our suffering	3.0	4.0	2.2
Compassionate attention	The process of becoming aware of or focus one's attention on things that are helpful and supportive	30.2	31.6	29.0
Compassionate reasoning	Thinking about and reflecting on the world, ourselves, and others from a balanced perspective with the purpose of alleviating or preventing suffering, and to experience these thoughts as kind, supportive, and helpful. This includes non‐judgement, that is, taking an accepting, non‐critical, and non‐condemning view/attitude towards ourselves and others	46.2	51.8	41.0
Compassionate behaviour	Engaging in activities/behaviours that help to alleviate or prevent suffering and facilitate positive feelings, development, and growth. This may involve both passive and active actions (e.g., using imagery/meditation to generate compassionate feelings for the self and others)	76.8	77.7	75.9
Compassionate feeling/sensation	Experiencing emotional and/or bodily feelings of warmth, kindness, and soothing when being compassionate, experiencing compassion from others or being self‐compassionate	35.0	42.5	28.1

I = improvers; NI = non‐improvers.

The percentages give the proportion of email messages wherein a particular attribute is expressed within the total number of sent email messages. For example, care for well‐being was expressed in 36.2% of all messages sent by improvers to their counsellor.

Table 3

Spearman intercorrelations of frequencies of compassionate attributes and skills

Compassionate attribute/skill	1	2	3	4	5	6	7	8	9
1. Care for well‐being	–
2. Sensitivity	.06	–
3. Distress tolerance	.09	.12	–
4. Empathy	.36*	.31*	.09	–
5. Common humanity	−.15	.06	−.02	.07	–
6. Compassionate attention	.02	.26*	.28*	.31*	.05	–
7. Compassionate reasoning	.21	.08	.16	.03	.02	.28*	–
8. Compassionate behaviour	.11	.24	.28*	.25	−.11	.30*	.03	–
9. Compassionate feeling/sensation	.18	.32*	.23	.36*	.09	.42***	.25	.32*	–

Benjamini–Hochberg false discovery rate‐corrected p‐values are used.

*p < .05; ***p < .001.

Multiple case study

To gain a better understanding of how the attributes and skills underlying compassion were manifested in the emails, a multiple case study approach was employed. Following the diverse case method (Seawright & Gerring, 2008), three illustrative cases with diverse backgrounds and motivations to participate in the study were drawn from the total sample. For the selection of cases, motivations for engagement in the intervention were primarily drawn from the first emails sent by the participants to their counsellor. The first and third author read through the emails of each case, repeatedly. Independently from one another, both authors took notes of their impressions and thoughts, thereby focusing on: the participant's background and motivation to follow the intervention, the participant's process of (re)constructing a compassionate self (e.g., which compassionate attributes/skills are helpful for the participant and how do these evolve over the course of the intervention), and outcomes of the intervention from the perspective of compassion. Next, the authors discussed the content of the emails, compared notes, and produced consensual descriptions of the three cases. In this paper, the cases are presented in narrative form, using fictitious names. The purpose of these qualitative case descriptions was not so much to present a complete picture of the participants’ experiences with compassion, but rather to provide a meaningful lens from which to understand the phenomenon of compassion. Each case exemplifies multiple compassionate attributes and skills, and altogether, the cases illustrate the range of deductively generated compassionate attributes and skills.

Quantitative analyses

To facilitate quantitative analyses, per participant and per attribute/skill an average frequency per message was calculated. First, the total frequency of the respective compassionate attribute/skill was determined by summing the number of counts per participant over the course of the intervention. Next, the total frequency of the attribute/skill was divided by the total number of email messages sent by the participant. From now on, when using the term ‘frequencies’, we refer to average frequencies.

We assessed intercorrelations across frequencies of compassionate attributes and skills using Spearman's rho (two‐sided, p < .05). Correlations were deemed weak (<.10), small (.10–.30), moderate (.30–.50), or strong (.50–1.00) (Cohen, 1988).

With regard to frequencies of compassionate attributes and skills, a comparison was made between improvers and non‐improvers in terms of well‐being. The RCT findings indicate that 42 out of the 87 eligible CFT participants (48.3%) showed clinically significant changes in well‐being from baseline to post‐intervention. These participants were therefore categorized as ‘improvers’. The remaining 45 participants (51.7%) did not exhibit clinically significant change and, hence, were labelled as ‘non‐improvers’. Improvers and non‐improvers did not significantly differ with regard to any of the demographics (see Table 1). Fisher's exact test (two‐sided) indicated no significant differences in study dropout (p = 1.000) and intervention adherence (p = 1.000) between improvers and non‐improvers. There was also no significant difference in the number of emails sent by improvers (n = 301, M = 7.17, SD = 0.66, range: 6–9) and non‐improvers (n = 324, M = 7.20, SD = 0.69, range: 5–9), t(85) = .23, p = .819. Independent samples t‐tests (p < .05) were conducted for each attribute/skill, separately.

Due to the large number of comparisons in the correlational analyses and t‐tests, the Benjamini–Hochberg procedure was employed to control the false discovery rate (FDR, i.e., expected ratio of n erroneous rejections to n rejections) (Benjamini & Hochberg, 1995; Thissen, Steinberg, & Kuang, 2002). The largest observed p‐value which was smaller than its FDR‐corrected p‐value and all p‐values smaller than that were deemed significant. Only FDR‐corrected p‐values are reported. The FDR was set at .05.

All quantitative analyses were conducted using SPSS version 24.0 (IBM Corp, Armonk, NY, USA).

Results

Testing the theoretical model of compassion

During the qualitative content analysis, several adaptations were made to the original theoretical model underlying CFT in order to better fit the data. First, given that sympathy was generally expressed simultaneously with empathy and rarely on its own (i.e., without empathy), the definition of empathy was broadened to a two‐dimensional attribute distinguishing between cognitive and emotional empathy. Second, based on the data at hand, reflecting on suffering of ourselves and others from a non‐judgemental view was considered a form of compassionate reasoning; hence, non‐judgement was integrated in the definition of compassionate reasoning. Third, it was found difficult to distinguish between emotional and bodily feelings of compassion, that is compassionate feeling and compassionate sensation. For instance, when participants reported feeling calm, content, grateful, or relaxed, it was often unclear whether this concerns physiological sensations such as a reduced heart rate or breathing, emotional sensations, or both. Therefore, compassionate feeling and compassionate sensation were merged into compassionate feeling/sensation. Fourth, in the context of the trial, for many participants, compassionate behaviour involved conducting exercises from the intervention. Since the intervention incorporated several visualization and meditation exercises, compassionate imagery may be considered a specific form of compassionate behaviour. Compassionate imagery was therefore integrated into compassionate behaviour. Finally, one additional compassionate attribute emerged from the emails, namely common humanity, which relates to the ability to recognize that suffering is part of the human experience and that we are not alone in our suffering.

Aforementioned modifications resulted in a slightly less nuanced model with five compassionate attributes and four compassionate skills, all of which were reflected in the emails. Conceptualizations of the compassionate attributes and skills are provided in Table 2.

For the attributes, sensitivity was most frequently expressed and common humanity the least frequent. The most and least frequently manifested skills were compassionate behaviour and compassionate attention, respectively. Participants showed on average 3.39 different attributes (SD = 0.83, range: 1–5) and 3.74 different skills (SD = 0.58, range: 1–4). Out of 87 participants, 4 (4.6%) exhibited all attributes and 69 (79.3%) exhibited all skills. Over the course of the intervention, three participants (3.4%) expressed all five attributes and all four skills in the emails.

Illustrative cases

To illustrate compassionate attributes and skills in a concrete context, we report three cases from the trial. Each case pertains to a different life domain: health, social relationships, and work, respectively.

Case 1: female, 53 years, compassion in the context of health

Emma experiences chronic stress and fatigue. As a breast cancer survivor, she experiences a lot of stress surrounding illness and death. Emma is familiar with mindfulness. As a next step, she wishes to develop more self‐compassion in order to be better able to deal with stress, hence her participation in the self‐help intervention [care for well‐being]. Over the course of the intervention, she becomes more aware of and attentive to her stress and underlying feelings of anxiety, guilt, anger, and sadness, but also to the suffering of others [sensitivity]. At first, she finds this confronting, but throughout the course she gradually learns to embrace and accept these unpleasant feelings [distress tolerance], leaving more room for feelings of calmness, contentment, and relatedness [compassionate feeling/sensation]. The self‐help book provides her with a lot of compassion resources which are increasingly integrated in her everyday life [compassionate behaviour], helping her to gain a feeling of control over her stress. A significant moment is in the fifth week when she has her annual cancer check‐up at the hospital. Emma describes how she applied several breathing and visualization exercises learned during the course [compassionate behaviour], enabling her to embrace feelings of restlessness, anxiety, and insecurity [distress tolerance] and reduce stress.

Case 2: male, 65 years, compassion in the context of relationships

Paul struggles with experiences of loss and grief at the start of the self‐help intervention, loss of health (prostate cancer), loss of job (recently retired), renewed grief over his deceased son, and most notably loss of his long‐term relationship. He reports mixed feelings of sadness, anger, regret, and revenge [sensitivity]. In the emails, he shows a change in perspective regarding his broken relationship, from a failure experience towards an enriching experience which enabled him to grow and become the person he is now [compassionate reasoning], which is accompanied by feelings of kindness and gratitude [compassionate feeling/sensation]. Not only does Paul adopt a more kind and accepting view towards his self, he is also able to expand his compassion to others, even to his ex‐girlfriend [compassionate reasoning]. As he lets go of his ex‐girlfriend and becomes increasingly at peace with the break‐up, he finds himself able to engage in and fully commit to existing and new meaningful friendships and relationships [compassionate behaviour]. Also, a shift in attention can be noticed from ‘losses’ to ‘gains’. In the end, Paul focuses more on all the positive things in his life and seems to have regained a sense of purpose and meaning [compassionate attention].

Case 3: female, 31 years, compassion in the context of work

From an early age, Lisa has been very self‐critical and vulnerable to stress. She has the tendency to set the bar too high, to focus on her shortcomings rather than on her strengths, and she does not allow herself to make mistakes. Central to her life story is the fear of not being good enough. Over the course of the intervention, Lisa becomes more and more aware of her self‐critical and self‐condemning thoughts [sensitivity]. She realizes that these unhealthy thought patterns can be traced back to early childhood experiences. She grew up in an unsafe neighbourhood, did not have a secure relationship with her mother, and experienced a lack of autonomy and parental support [empathy]. Lisa has the deep desire to feel more content with herself, and hopes to achieve this by participating in the self‐help intervention [care for well‐being]. From the emails, it becomes clear that especially in the work domain she feels stressed, frustrated, and insecure, due to a negative and uninspiring atmosphere among colleagues [sensitivity]. The exercises contribute to a sense of empowerment. Step by step, Lisa gains the courage and confidence to undertake actions towards altering her work situation. Before the end of the course, she takes several major steps, for example applying for a job and exploring opportunities to start her own business [compassionate behaviour].

Intercorrelations between compassionate attributes and skills

Intercorrelations of (frequencies of) compassionate attributes and skills are shown in Table 3. In general, most correlations were positive but insignificant. All significant correlations fell in the small to moderate range. It appears that compassionate skills are more intercorrelated than compassionate attributes. Only common humanity was not correlated with any of the other attributes or skills. Most correlations were found for compassionate attention, followed by empathy and compassionate feeling/sensation.

A comparison of improvers and non‐improvers

Compassionate attributes

All attributes were expressed in the emails of both improvers and non‐improvers. For both improvers and non‐improvers, sensitivity was the most commonly expressed compassionate attribute, followed by empathy (see Table 2). Common humanity was the least observed attribute in the email narratives, regardless of whether one profited from the intervention or not. None of the compassionate attributes yielded statistically significant differences in frequencies across improvers and non‐improvers (p ≥ .297). Also no statistically significant differences were found in the number of different attributes between improvers and non‐improvers, t(85) = .37, p = .716. Improvers and non‐improvers expressed minimum one and maximum five different attributes (improvers: M = 3.36, SD = 0.82; non‐improvers: M = 3.42, SD = 0.84).

Compassionate skills

All four compassionate skills emerged in the emails of improvers and non‐improvers. Compassionate behaviour was most frequently observed in the emails of both groups, that is, in more than three‐quarters of all sent messages. The lowest‐frequency skill for the improvers was compassionate attention. For non‐improvers, the lowest‐frequency skill was compassionate feeling/sensation, closely followed by compassionate attention. Compassionate feeling/sensation was more frequently expressed by improvers compared to non‐improvers during the 9‐week intervention period, t(85) = −3.14, p = .018. For the remaining skills, no significant differences in frequencies were found between improvers and non‐improvers (p ≥ .171). The diversity in compassionate skills did not significantly differ across groups, t(75.7) = −1.55, p = .125, with improvers showing 2–4 different skills (M = 3.83, SD = 0.44) and non‐improvers showing 1–4 different skills (M = 3.64, SD = 0.68).

Discussion

To our knowledge, this is the first study that empirically explored components of compassion as outlined in the theoretical model underpinning CFT. In the intervention group of a recent RCT, who followed a CFT‐based self‐help intervention with email counselling for enhancing well‐being, we sought support for the model through analysing email narratives from participants to their counsellor. While the model presumes that compassion involves six attributes and six skills, supposedly theoretical distinct, during the coding process it proved difficult to empirically distinguish all these attributes and skills. We slightly modified the original model to better fit the data, resulting in a simplified model including five compassionate attributes (i.e., care for well‐being, sensitivity, empathy, distress tolerance, and common humanity) and four compassionate skills (i.e., compassionate attention, reasoning, behaviour, and feeling). All of these attributes and skills emerged in the emails. As such, our data largely support the theoretical model of compassion underlying CFT.

Whereas the five compassionate attributes identified in the current study have been acknowledged as facets of compassion in prior research, it seems that transformative skills of compassion have been largely ignored in existing definitions and measures of compassion (Strauss et al., 2016). In this light, an interesting observation in the current study was that a large proportion of participants expressed all four compassionate skills (79.3%), albeit only a few (n = 4, 4.6%) exhibited all five compassionate attributes. A potential explanation is that learning compassionate attributes takes place more implicitly, that is, outside people's awareness, while learning compassionate skills takes place more explicitly or inside people's awareness, leading participants to express skills more frequently than attributes. It is also possible that participants were less inclined to write about engagement attributes, because the intervention was more action‐oriented; that is, emphasis was on skills of compassion.

Our findings seem to suggest that, in the process of (re)constructing a compassionate self, some attributes and skills are more central than others. For the attributes, most outstanding were sensitivity, the ability to recognize suffering, and empathy, the ability to feel sympathy or empathy for suffering. Of all compassionate attributes, these two emerged most frequently in the emails, which may be partly attributed to the email counselling. Writing emails to their counsellor gave participants the opportunity to review and reflect on their problems and struggles in everyday life. For sensitivity and empathy, a positive association was found with one or two attributes as well as with some of the compassionate skills, that is compassionate attention and compassionate feeling/sensation. As such, sensitivity and empathy seem to function as hubs connecting the two mindsets of compassion. In the theoretical model, cognitive empathy is proposed as a higher‐order attribute nourished by sensitivity, sympathy (i.e., emotional empathy), and distress tolerance. Being able to engage with and tolerate suffering is considered a prerequisite for being capable of having empathic insights (Gilbert, 2014a, 2014b). Our data confirm a link with sensitivity, but not with distress tolerance.

With regard to compassionate skills, compassionate behaviour was by far the most frequently reported skill. This is not surprising as practising exercises from the self‐help intervention was coded as compassionate behaviour. Whereas some participants may have integrated the exercises and principles from the intervention in their daily lives, resulting in enduring behavioural changes, for others these may have been temporary behaviours which dissipated after the end of the intervention. Of greater interest are compassionate attention and compassionate feeling/sensation. Compassionate attention, being the only skill that showed a positive correlation with each of the other skills, seems to be at the heart of the second mindset of compassion. This concept may be regarded as the positive counterpart of sensitivity. It seems plausible that being attentive to helpful others, experiences or knowledge encourages participants to think, act, and feel more compassionate towards oneself and others.

In relation to well‐being, compassionate feeling/sensation bears particular interest. The comparison of improvers and non‐improvers revealed that improvers expressed significantly more compassionate feeling/sensation than non‐improvers, while no significant differences were observed for the remaining attributes and skills. These findings suggest that CFT may instigate well‐being if a client is able to experience compassionate feelings for oneself and others. This suits with what is denoted as the central aim in CFT, namely to strengthen individuals’ affect regulation system of soothing and affiliation, enabling individuals to reassure and soothe oneself (Gilbert, 2009, 2014b). It is possible that non‐improvers experienced difficulties accessing the soothing and affiliation system, hence, did not feel safe, secure, or content when engaging in compassionate thoughts or practicing compassionate behaviours. Previous research affirms that perceived cognitive changes do not necessarily result in affective changes, which is commonly referred to as the cognition–emotion mismatch (Stott, 2007). Thus, when generating helpful thoughts in CFT, it is important that participants not only understand the logic behind these thoughts but also emotionally experience these thoughts as kind, reassuring, and soothing (Gilbert, 2009, 2014a). An alternative explanation for the finding that improvers express more compassionate feeling/sensation than non‐improvers is that improvers either felt more comfortable in disclosing feelings of compassion or were more capable in giving words to their feelings (Barak, Klein, & Proudfoot, 2009; Richards & Vigano, 2013). Data from the trial (Sommers‐Spijkerman et al., 2018aa) indicate that improvers exhibited significantly higher levels of self‐reassurance and well‐being beforehand, as compared to non‐improvers. This may suggest that improvers were already better able to access emotional states of soothing and affiliation, leading them to express these emotions more regularly in the emails.

Whereas compassionate feeling/sensation appears to matter most for improving well‐being, other attributes and/or skills may be more important for decreasing distress. For instance, it seems reasonable that distress tolerance and common humanity, compassionate attributes which were infrequently used by our sample, are more relevant to (clinical) samples with higher levels of distress. In patients with psychosis, Waite et al. (2015) found that acceptance of the self and acceptance of psychosis as well as the distress associated with it (e.g., shame) are central to their recovery process, suggesting that non‐judgement and distress tolerance are major compassionate attributes in this specific clinical population. Lawrence and Lee (2014) found that among others the experience of common humanity is crucial to PTSD patients’ process of developing self‐compassion. The awareness that other patients had similar struggles led PTSD patients to accept that these were understandable responses to their experiences.

Strengths and limitations

A major strength of the study is that no fixed format was used for the email guidance, meaning that the emails reveal spontaneous self‐disclosures of compassion. In this regard, it seems plausible that if one expressed a particular attribute or skill more often, one actually experienced or practised it more frequently, at least consciously.

Nonetheless, there are notable limitations to this study. First, some of the used definitions in the adapted model of compassion may elicit discussion among researchers and practitioners. It may for instance be argued that sympathy and empathy are rather different processes that should not have been combined in the coding scheme, especially given previous work in this area (e.g., Loewenstein & Small, 2007; Wispé, 1986). However, as our primary aim was to empirically test Paul Gilbert's multidimensional model of compassion, we followed the definitions of Gilbert (Gilbert, 2009, 2014b) who approaches sympathy and empathy as two dimensions of relating to suffering. Second, compassionate attributes and skills were trained implicitly rather than explicitly. As the self‐help book does not explicitly mention the attributes and skills of compassion, participants wrote about their personal experiences in laymen's terms. Consequently, we interpreted which attributes and skills were reflected in the contents of the emails. Third, we captured what the participants wrote in the emails rather than what they did in CFT. A fourth limitation is that we were only able to explore compassionate attributes and skills in the intervention group of the RCT and not in the waitlist control group who received CFT without email guidance. Because the process of writing emails to a counsellor may be therapeutic in itself (Richards & Vigano, 2013), we cannot rule out that (some of the) compassionate attributes and skills, as revealed in the content analysis, were cultivated by the email counselling rather than by the contents of the CFT intervention. In favour of this possibility, our RCT findings suggest that the self‐help intervention is more effective in improving self‐compassion when offered with email guidance. Fifth, we did not code how participants responded to the emails from the counsellor. Investigating the interaction between participants and counsellors may be informative as well. Sixth, we used a self‐selected non‐clinical sample with low to moderate levels of well‐being and low levels of distress. As such, our findings may not be generalizable to clinical samples with more serious levels of distress. Different problems/symptomatologies or motives may require different sets of compassionate attributes and skills. Finally, inter‐rater reliability was sufficient but not optimal, though disagreements were resolved by discussion.

Future directions

As our primary aim was to find preliminary support for the theoretical model underlying CFT, we used a deductive rather than an inductive approach when analysing the content of the emails. We recognize that additional attributes and skills may be involved that we are currently not aware of, hence encourage qualitative research in this area. Qualitative research wherein participants are explicitly asked about their experiences during CFT and helpful or hindering factors, for example using the change interview developed by Elliott (2008, 2012), may reveal key processes that were not described in the emails.

In contrast to the theoretical notion that all compassionate attributes and skills are interdependent (Gilbert, 2014b), correlations across (frequencies of) compassionate attributes and skills were mostly low and insignificant. Although this may suggest that the attributes and skills reflect sufficiently distinct facets of compassion, stronger intercorrelations may be expected from components of a theory. Hence, the model may require further refinement and testing as to broaden its scientific applicability.

More work is needed to establish the theoretical and empirical distinctiveness of the identified compassionate attributes and skills. Following replication in future studies, these attributes and skills may form the basis for developing a more comprehensive measure of compassion.

Despite that our findings cast doubt upon the usefulness of a number of attributes and skills outlined in Gilbert's model (Gilbert, 2009, 2014b) when applied to a non‐clinical population (e.g., distress tolerance), future research may establish its applicability in clinical samples. From a more practical perspective, it would be useful to assess which compassionate attributes and skills can be cultivated by which CFT exercises.

Whereas the specific aim of our paper was to seek empirical support for attributes and skills underlying compassion, as pre‐defined by Gilbert (2009, 2014b), for future research, the three qualities of courage, commitment, and wisdom central in CFT may also offer a meaningful lens when viewing participants’ responses to CFT.

Conclusion

In conclusion, we did find that the dual model of compassion underlying CFT is feasible and useful, albeit our data suggest that the set of attributes and skills underlying these two dimensions may be too nuanced for application in a non‐clinical population. Furthermore, our data suggest that compassionate feeling/sensation may be of special interest when well‐being is the intended outcome of therapy.