Practicing Compassion Increases Happiness and Self-Esteem
The current study examined the effect of practicing compassion towards others over a 1 week period. Participants (N = 719) were recruited online, and were assigned to a compassionate action condition or a control condition which involved writing about an early memory. Multilevel modeling revealed that those in the compassionate action condition showed sustained gains in happiness (SHI; Seligman et al. in Am Psychol 60:410–421, 2005) and self-esteem (RSES; Rosenberg in Society and the adolescent self-image. Princeton University Press, Princeton, 1965) over 6 months, relative to those in the control condition. Furthermore, a multiple regression indicated that anxiously attached individuals (ECR; Brennan et al. 1998) in the compassionate action condition reported greater decreases in depressive symptoms following the exercise period. These results suggest that practicing compassion can provide lasting improvements in happiness and selfesteem, and may be beneficial for anxious individuals in the short run.
KeywordsCompassion Compassionate action Attachment Happiness Self-esteem
Compassion is often considered an important human strength, requiring a sense of caring, empathy, and sympathy, each of which enable one to connect with and care for another. Of notable relevance to mental health, compassion is not only a process that builds positive relationships with others; it is also a vital path to releasing the human mind from the effects of harmful negative emotions (Wang 2005).
Despite the centrality of compassionate attitudes and actions in Eastern traditions, compassion has only recently garnered attention as a recognized psychological construct (Gillath et al. 2005). The reciprocity of compassionate behaviour has also often been neglected, with most studies demonstrating the health benefits for recipients of compassionate support (e.g., Krokavcova et al. 2008; Smith and Howard 2008), rather than those accrued by the provider of support. Thus, both theory (e.g., Brown and Brown 2006) and empirical evidence highlight the need to further examine the experience of the provider of compassion in a psychological health context.
While some studies dispute the psychological benefits of compassion-rooted behaviours (e.g., O’Malley and Andrews 1983), a converging body of literature suggests that various forms of compassionate behaviours have immediate and long-term psychological health benefits for the individuals who exhibit them, including positive mood states (Millar et al. 1988), reduced depressive symptoms (Krause et al. 1992; Taylor and Turner 2001), and increased self-esteem (Krause and Shaw 2000; Yogev and Ronen 1982). Compassion-based behaviours may even provide a coping mechanism for stressful life events and transitions. For example, one study found that individuals with a greater number of compassionate goals experienced increased social support (Crocker and Canevello 2008).
Furthermore, there is mounting evidence demonstrating an association between psychological gains and prolonged helping behaviours performed within institutional settings, namely volunteering. For example, in a correlational study by Thoits and Hewitt (2001), number of hours of volunteer work was found to predict positive changes in well-being, happiness, life satisfaction, self-esteem, mastery, depression, and physical health over the subsequent 3 years.
Perhaps the most compelling evidence originates from studies demonstrating the superior benefits of the provision of compassion over the receipt of compassionate support. For instance, a correlational study conducted by Schwartz et al. (2003) found that being the provider of altruistic behaviours was associated with greater gains in mental health (i.e. lowered depression and anxiety) than being the recipient of altruistic behaviours. Additionally, Dunn et al. (2008) found that individuals who were randomly assigned to spend a windfall of money on others exhibited greater gains in happiness than those instructed to spend a windfall on themselves.
In Buddhist traditions, compassionate orientations are typically promoted through the use of imagery and meditation procedures (Allen and Knight 2005). One such practice is referred to as loving-kindness meditation. Its objective is to effect positive cognitive, emotional, and behavioural changes by training individuals to impart compassion and goodwill toward “real or imagined others” (Hutcherson et al. 2008, p. 720). Hutcherson et al. (2008) conducted one of the few studies to empirically test the efficacy of loving-kindness meditation in eliciting changes in mood. Participants were trained in a 7-min visualization procedure, in which they were to imagine two loved ones sending loving and compassionate feelings towards oneself, and to subsequently redirect these warm feelings toward the photograph of an emotionally neutral stranger (Hutcherson et al. 2008). Participants then recited a succession of phrases wishing the stranger health, happiness, and well-being. Relative to the control group, participants who engaged in this loving-kindness visualization experienced more positive moods and less negative moods, in addition to more positive affective responses to the self and others. These findings suggest that even a brief procedure designed to cultivate compassion towards others may have upward effects on mood (Hutcherson et al. 2008).
Recently, Lyubomirsky et al. (2005) reported data describing the use of a 6-week positive psychology intervention designed to increase happiness and leverage one’s personal strength of kindness. Over the course of 6 weeks, participants were randomly assigned to complete five acts of kindness that would benefit others—either all in 1 day or spread throughout the week. These acts varied in terms of personal costs, ranging from donating blood to writing a thank-you note to a former professor. The control condition consisted of withholding treatment, with participants only completing measures of well-being. Results were promising, showing that individuals who performed five acts of kindness per week exhibited gains in well-being over the study period, providing that all acts were performed in a single day. On the contrary, control participants experienced declines in happiness.
1.1 Matching Compassion to Attachment Style
The need to assess the ways in which attachment styles differentially influence the response to a compassion exercise is timely. Lyubomirsky et al. (2005) maintain that exercise efficacy can be maximized by matching people to intentional therapeutic activities according to their unique personalities, needs, values, and preferences. This postulation is supported by several studies by Fordyce (1977, 1983) which found that happiness-increasing activities were effective to the extent that they corresponded to the needs and weaknesses of the individual (Lyubomirsky et al. 2005). Given the interpersonal-oriented nature of compassionate acts, attachment style appeared to be a particularly relevant to the study of individual differences in the benefits obtained from a compassion-based exercise.
In line with Bowlby’s (1973) accentuation on the continuity of the attachment system across the life span, numerous theoretical models have endeavored to describe attachment styles in adulthood. One model, proposed by Bartholomew (1990) delineates four adult attachment patterns comprised of combinations of abstract images of self (loveable vs. unlovable) and others (trustworthy, accepting and reliable vs. untrustworthy and rejecting) (Bartholomew 1990; Bartholomew and Horowitz 1991). In the context of romantic relationships, Brennan et al. (1998) conceptualized attachment styles among adults as existing along two orthogonal dimensions. The first dimension, anxiety, indicates the extent to which an individual worries about abandonment and rejection by significant others (Brennan et al. 1998).
The second dimension, avoidance, indicates the degree to which an individual has difficulty in achieving intimacy with others and struggles to preserve a sense of autonomy and self-reliance. Avoidant individuals adopt deactivating strategies involving the dismissal of attachment-related needs and avoidance of intimate interactions (Mikulincer and Shaver 2007).
Individuals scoring low on both anxiety and avoidance are believed to possess a secure attachment style. Secure individuals attempt to foster interdependent relationships, in which a balance between intimacy and autonomy is realized (Collins and Read 1990; Hazan and Shaver 1987).
The quality of our interpersonal relationships is a significant determinant of our psychological health and sense of self-worth (see McAdams and Bryant 1987 for a review). This is particularly true for anxiously attached individuals whose psychological functioning and well-being are profoundly entwined with interpersonal concerns. It is therefore possible that such persons may derive a number of benefits from compassionate goals, including an increase in the amount of social support received (Crocker and Canevello 2008). Overall, incorporating compassion into one’s interpersonal activities may ease anxious individuals’ concerns regarding the availability, reliability, and responsiveness of others by creating a supportive social milieu in which one’s own requests for support are more likely to be reciprocated.
The integration of compassionate acts into one’s social behaviour repertoire may also initiate a multitude of self-validating and self-esteem enhancing experiences. Anxious individuals may begin to value themselves as compassionate beings who play an important role in the lives of others, thereby increasing their depleted feelings of self-worth and mattering—both of which have been shown to enhance well-being (e.g., Roberts et al. 1996; Taylor and Turner 2001). By becoming the provider rather than compulsive seeker of compassionate support, anxious individuals could also derive a sense of empowerment and self-reliance.
The defensive strategies of avoidant individuals may preclude the possibility of deriving particularly strong mental health gains from the proposed intervention. Avoidant attachment has been shown to entail fewer compassionate goals (Crocker and Canevello 2008) and the interpersonal nature of the compassionate action exercise may be perceived as objectionable. As discussed previously, highly avoidant individuals cope with attachment insecurity by avoiding romantic situations that activate the attachment behavioural system (Mikulincer et al. 2003). Although the proposed exercise was not geared towards romantic partners, it may nonetheless necessitate a level of closeness that may unwelcomingly challenge the avoidant person’s active attempts to maintain interpersonal distance (Mikulincer et al. 2003). For those reasons, a compassionate action exercise was not expected to provide particular benefits for those with an avoidant attachment style.
1.2 Overview and Hypotheses
The present work involved an online study in which participants were randomly assigned to a compassionate action condition, or a control group. Participants in the compassion group were required to interact with someone in a supportive and caring way on a daily basis for 1 week. The compassionate action exercise was compared to an early memories control condition in which participants were to write a detailed description of an early memory, also on a daily basis for 1 week (see Seligman et al. 2005 for a similar procedure).
All participants completed self-report measures of depression, happiness, self-esteem, and attachment at the onset of the study. Well-being, comprised of depression, happiness, and self-esteem, was assessed at four subsequent time points over a 6 month period. This longitudinal study was thus comprised of a 1 week exercise period followed by post-experimental assessments for up to 6 months.
The central hypothesis was as follows: Compared to writing about an early memory, performing daily compassionate actions for 1 week should result in greater reductions in depression and greater increases in happiness and self-esteem over the 6 month period.
An additional goal of this study was to examine potential interactions between attachment and responses to the compassion exercise. It was expected that those high in anxious attachment would experience greater reductions in depressive symptoms and greater increases in happiness and self-esteem in the compassionate action condition. Those with an avoidant attachment style were not expected to show superior gains from the compassion exercise and this attachment style was not expected to be a significant predictor of outcome.
Participants were recruited via the Internet through advertisements placed on Facebook, inviting individuals to participate in an online study by York University researchers that explored the effect of positive exercises on mood. The initial sample consisted of 719 participants ranging in age from 17 to 72 years (M = 33.63, SD = 11.52). There were 591 (82.2%) females and 118 (16.4%) males, and 10 (1.4%) participants did not respond to this question. No exclusionary criteria, other than being Canadian and over 18 years of age were employed for this study.1
Overall, the sample was predominantly Christian (48.5%) and of Caucasian heritage (79.4%). The remaining participants identified themselves as Asian (5.8%), Mixed Heritage (4.3%), East Indian (2.6%), Black (2.1%), Middle Eastern (1.5%), “Other” (1.5%), Aboriginal or Inuit (1.4%), and Hispanic (1.0%). Participants were highly educated overall, with 82.7% reporting they had completed at least some college education. Additionally, the majority of participants (74.9%) had an annual income of less than $50,000, with 2.5% not responding to this question. With respect to marital status, 37.4% of the participants were single and did not have children (55.1%). Of the remaining individuals in the study, 28.9% were married, 17.7% were currently dating, and 15.7% were co-habitating.
2.1.1 Participant Remuneration
Participants entering the study before November 1st, 2007 received a total of 30 dollars in remuneration upon completion of the 1 week exercise and measures. In the second round of data collection, participants were not offered 30 dollars remuneration, however were entered into a $1000 prize draw upon completion of the 1 week exercise and assessments. All participants were entered into a $1000 draw following completion of each follow-up at 1, 3, and 6 months. Winners were informed by telephone.
2.2.1 Demographic Questionnaire
An author-constructed demographics survey was administered at the beginning of the study, similar to the one administered during an online study at http://www.authentichappiness.org. This questionnaire assessed demographic characteristics of participants including age, gender, race and ethnicity, citizenship, religion, marital status, number of children, highest level of educational attainment, and income range.
2.2.2 Steen Happiness Index (SHI; Seligman et al. 2005)
The SHI consists of 20 items tapping levels of happiness over the previous week. Its items are based on Seligman’s (2002) theory which states that happiness is determined by three core elements: pleasure, engagement and meaning (Seligman et al. 2006). Participants are required to select one of five statements, ranging from negative to positive, that most describes their present state. An example includes response choices ranging from 1 (life is bad) to 5 (life is wonderful). To demonstrate convergent validity in non-clinical adult populations, Seligman et al. (2005) found the SHI to correlate highly with other measures of happiness, including the General Happiness Scale (Lyubomirsky and Lepper 1999; r = .79) and the Happiness Scale (Fordyce 1977; r = .74).
2.2.3 Center for Epidemiological Studies Depression Scale (CES-D; Radloff 1977)
The CES-D is a 20-item self-report scale that assesses the frequency of depressive symptoms, such as hopelessness and depressed mood, within the previous week. Responses are ranked on a 4-point Likert scale, ranging from 0 (rarely or none of the time) to 4 (most or all of the time). Sample items include “I was bothered by things that usually don’t bother me,” or “I felt that everything I did was an effort.” The scale has been shown to have internal consistency of .8 or higher (Klinedinst et al. 2007) and split-half reliability coefficients ranging from .77 to .92 (Corcoran and Fisher 1987). The CES-D has good validity, being strongly correlated with other commonly used measures of depression (Santor et al. 1995), and highly discriminating of individual differences in symptom severity (Santor et al. 1997).
2.2.4 Rosenberg Self-Esteem Scale (RSES; Rosenberg 1965)
The RSES is a self-report unidimensional measure of global self-esteem. It consists of 10 statements related to overall feelings of self- worth, respect and confidence. Examples of such items are “On the whole, I am satisfied with myself” or “I feel that I’m a person of worth, at least on an equal plane with others.” Responses are ranked on a four-point scale ranging from strongly agree to strongly disagree. Scores range from 10 to 40 with higher scores reflecting higher self-esteem. Construct validity and concurrent validity have been demonstrated (Carmines and Zeller 1979). Good internal consistency has also been found, with Cronbach’s alpha coefficients ranging from .83 (Yarcheski and Mahon, 1989) to .99 (Damji et al. 1996). The RSES reportedly has good test–retest reliability (Fleming and Courtney 1984).
2.2.5 Experiences in Close Relationships (ECR; Brennan et al. 1998)
This self-report scale consists of 36-items used to measure adult romantic attachment. Respondents are required to indicate the extent to which each item is descriptive of their feelings in close relationships, on a 7-point scale ranging from strongly disagree (1) to strongly agree (7). Items were derived from a factor analysis of numerous self-report measures of adult romantic attachment. Factor analysis of the ECR yielded a two-dimensional, orthogonal factor structure: an Avoidance dimension, indicative of discomfort with interpersonal closeness and fear of intimacy, and an Anxiety dimension, indicative of fear of interpersonal rejection and preoccupation with abandonment (Brennan et al. 1998). The Avoidance subscale consists of 18 items such as “I prefer not to be too close to romantic partners” and “I find it difficult to allow myself to depend on romantic partners.” The Anxiety subscale consists of 18 items such as “My desire to be very close sometimes scares people away” and “I resent it when my partner spends time away from me.” Brennan et al. (1998) reported a high level of internal consistency with coefficient alphas of .91 and .94 for the Anxiety and Avoidance subscales, respectively. Test–retest reliabilities for the ECR are reportedly between .68 and .71 for both subscales (Lopez and Gormley 2002). In terms of validity, numerous studies have supported the convergent and divergent validity of the ECR with respect to both scales (e.g., Fairchild and Finney 2006; Wei et al. 2006).
Advertisements placed on Facebook invited individuals to participate in a study entitled “Project HOPE: Harnessing One’s Personal Excellence,” examining the effects of mental exercises on mood and well-being. The ad for the project appeared to Canadian Facebook users over 18 years of age. Once they clicked on the ad, prospective participants were directed to the project website (http://www.ProjectHopeCanada.com). All participants who registered for the study were first asked to complete a battery of measures online, including a demographic questionnaire, the SHI, CES-D, RSES, and ECR. Following completion of these baseline measures, participants were randomly assigned to the compassionate action condition or the early memory control condition. Instructions were provided describing their daily exercise and the need to report to the website each night to complete the exercise.
Participants in the active condition were subsequently asked to act compassionately towards someone for 5–15 min the following day, by actively helping or interacting with someone in a supportive and considerate way. Several examples of compassionate actions were offered, including “talking to a homeless person” and “simply being more loving to those around you.” During the evening of their daily compassionate act, participants were instructed to log onto the website to report about their experience.
Participants in the early memories control condition were asked to engage in a daily psychological exercise consisting of describing an early memory. Participants were subsequently asked to spend 10 min each night writing about an early memory in as much detail as possible. They were asked to describe this early memory in terms of what they were doing, what they were feeling, and by whom they were accompanied. An unlimited amount of space was provided for each online entry.
At the conclusion of the 7 day study period, participants in both conditions were asked to complete the SHI, CES-D, and SES, which served as outcome measures. The first 33.7% of participants who were paid at this point, while the remaining participants were entered into a $1000 draw.
At 1 month following the exercise period of the study, e-mails were sent to participants encouraging them to return to the website to complete a battery of follow-up questionnaires. These again consisted of the SHI, CES-D, and SES, as well as questions assessing their continued practice of the exercise. At 2 months after post-test, reminder e-mails were sent encouraging participants to continue their assigned exercise if they found it effective. Finally, e-mails were sent at 3 and 6 months post test reminding participants to return to the website to complete the outcome measures. Participants were entered into a separate 1,000 dollar prize raffle following completion of each follow-up assessment.
Correlations among study variables at baseline and 1 week (N = 719)
4. Anxiety(n = 712)b
5. Avoidance (n = 712)c
6. Age (n = 704)
7. Income (n = 694)
8. Self-esteem (Baseline)d
9. Self-esteem (n = 469) week)e
10. Happiness (baseline)f
11. Happiness (n = 468) (1 week)g
12. Depressive symptoms (baseline)h
13. Depressive symptoms (n = 471) (1 week)
$20 K –$30 K
3.2 Completers vs. Non-Completers Across Time
Mean scores (and standard errors) for depressive symptoms, happiness, and self-esteem for each condition over time
Interestingly, the condition assigned to participants predicted completion of the project at 1 week and 1 month. Those assigned to the compassionate action condition were more likely to drop out at 1 week, (t(717) = 2.41, p = .02, all 2-tailed), and at 1 month, (t(717) = 2.75, p = .006). This may be attributed to the onerous nature of the exercise, compared to the early memories condition which did not require any interaction with others.
Age was related to project completion at all time points, with younger participants being more likely to drop out at 1 week, 1, 3 and 6 months (ts(708) < −2.76, p < .01). 2 Men were more likely to drop out at 1 month and 6 months (ts(707) > 2.14, p < .05). Less educated participants were more likely to drop out at 3 and 6 months (ts(717) < −2.58, p = .01).
Payment status (whether participants were paid or entered in a lottery) influenced completion of the study, with those being paid being more likely to complete the project at all time points (ts(717) < −7.80, p < .001). Getting paid may have created a sense of obligation to complete the follow-up questionnaires. Adherence to the exercise regimen was similarly related to project completion. Those who completed their “homework” were more likely to complete the follow-up measures at all time points (ts(717) < −11.30, p < .001). Finally, those with higher self-esteem were more likely to complete the project at 1, 3, and 6 months (ts(717) < −2.07, p < .05). Attachment styles were unrelated to the likely hood of dropping out of the study.
In summary, those who were younger, less educated, male, not paid, low in self-esteem and who completed fewer of their exercises were more likely to drop out during the 6 month project.3
3.3 Exercise Effectiveness Using Multi-Level Modeling
Given the high attrition rate in our sample, a data analytic approach capable of handling missing data was adopted. Multi-level models were run on participants who provided responses at baseline and at 1 week (n = 458), in order to preserve a more balanced data set and derive more reliable regression estimates. Maximum likelihood estimation was used to estimate parameters in the models.
These analyses were conducted to determine the effect of the exercises on well-being over time, and were performed on 3 outcome measures including self-esteem, happiness and depression. Payment status, age, income, and adherence were entered as fixed effects, and controlled for in all models. The rate of change in self-esteem, happiness, and depression over the 6 month period are described next.
The model tested included the control variables (payment status, age, income, and adherence), along with exercise condition, followed by all interaction terms with Time. Inspection of the random effects for this model revealed that the rate of change among participants, or the slope of the trajectories, did not vary significantly (Estimate = .001, SE = .001, z = .93, p = .18) while controlling for all variables in the model. As well, participants’ initial status in terms of self-esteem did not show significant variability (Estimate = .0001, SE = .005, z = .03, p = .98). Furthermore, the rate of change for participants did not covary with their baseline levels of self-esteem (Estimate = −.59, SE = 1.25 z = −.47, p = .64), suggesting that participants’ self-esteem status at the outset of the study did not systematically impact their rate of change over time.
In terms of the fixed effects, a Time by condition interaction was obtained (Estimate = .04, SE = .02, t = 2.75, p = .006). Inspection of the estimates indicated that participants in the compassion condition showed greater increases in self-esteem than those in the control condition.4 (The control condition was the reference group for interpreting the fixed effects). Therefore, support for our hypotheses was obtained in demonstrating the superiority of the compassion exercise in improving self-esteem over time.
The model for happiness included as fixed effects: income, adherence, payment status, age, exercise condition, as well as the interaction between Time and all variables. Inspection of the random effects for this model revealed that the rate of change (Estimate = .003, SE = .001, z = 3.30, p = .005) and baseline values for happiness (Estimate = .47, SE = .04, z = 13.59, p < .0001) showed significant variability across participants. However, the rate of change for participants did not covary with their baseline levels of happiness (Estimate = −.004, SE = .005 z = −.73, p = .47). This indicates that participants’ happiness scores at the outset of the study did not systematically impact their trajectory or rate of change over time.
A Time by condition interaction effect was obtained (Estimate = .04, SE = .01, t = 3.27, p = .001), and inspection of the estimates indicated that those in the compassion condition experienced greater increases in happiness than those in the control group (reference group).5 In short, participants assigned to the compassion exercise experienced greater gains in happiness over the 6 month period.6
As for the previous two models, fixed effects included income, adherence, payment status, age, exercise condition, as well as the interaction between Time and all variables. Inspection of the random effects revealed that the rate of change (Estimate = .79, SE = .44, z = 1.82, p = .04) and initial status in terms of depressed symptoms (Estimate = 110.80, SE = 9.33, z = 11.87, p < .0001) showed significant variability across participants. However, the rate of change for participants did not covary with their baseline levels of depression (Estimate = -2.65, SE = 1.74, z = −1.52, p = .13). This indicates that participants’ levels of depression at the outset of the study did not systematically impact their trajectory or rate of change over time.
Inspection of the fixed effects failed to reveal a significant Time by condition interaction in the prediction of depressed symptoms (Estimate = −.34, SE = .27, z = −1.24, p = .21). There was a marginal effect for Time with reference to the compassion condition (Estimate = −.37, SE = .20, z = −1.90, p = .06). The effect for Time was nonsignificant for the control group (Estimate = −.03, SE = .19, z = −.18, p = .86). This suggests the compassion exercise lowered depression levels, while the control group did not. However, the effects here were more modest than those obtained for happiness and self-esteem and failed to produce a Time by condition interaction effect.7
3.7 The Influence of Attachment on Well-Being
The subsequent analyses included attachment styles as predictors of change following the exercise period. Our hypotheses pertained to the differential effects of the experimental condition for anxious and avoidant participants, so we were particularly interested in the Time by Condition by Attachment interaction effects in predicting changes in well-being.
The model tested included the control variables (income, adherence, payment status, age) along with the exercise condition, followed by the attachment variables and interaction terms predicting self-esteem over 6 months. Inspection of the random effects for this model revealed that the rate of change among participants, or the slope of the trajectories, did not vary significantly (Estimate = .002, SE = .001, z = 1.33, p = .09) while controlling for all variables in the model. But participants’ initial levels of self-esteem did show significant variability (Estimate = .20, SE = .02, z = 10.50, p < .0001). Furthermore, the rate of change for participants did not covary with their baseline levels of self-esteem (Estimate = −.003, SE = .004, z = −.61, p = .54), suggesting that participants’ self-esteem status at the outset of the study did not systematically impact their rate of change over time.
In terms of the fixed effects, main effects for Avoidance (Estimate = −.17, SE = .03, t = −6.10, p < .001) and Anxiety (Estimate = −.14, SE = .03, t = −5.43, p = . < .001) were obtained. Inspection of the estimates indicated that participants who were more anxious or avoidant were lower in self-esteem at baseline.8 None of the three-way interactions were significant, contrary to our predictions. That is, the effect of the compassionate exercise on self-esteem was not found to vary according to attachment style.
The model tested included the control variables (income, adherence, payment status, age) along with the exercise condition, followed by the attachment variables and interaction terms predicting happiness over 6 months. Inspection of the random effects for this model revealed that the rate of change among participants, or the slope of the trajectories, varied significantly (Estimate = .003, SE < .001, z = 3.15, p = .0008) while controlling for all variables in the model. Initial status, or participants’ baseline values in happiness did show significant variability (Estimate = .37, SE = .03, z = 13.19, p < .0001). However, the rate of change for participants did not covary with their baseline happiness values (Estimate = .0009, SE = .004, z = .23, p = .82), indicating that participants’ happiness at the outset of the study did not systematically impact their rate of change over time.
Main effects for Avoidance (Estimate = −.18, SE = .04, t = −4.92, p < .001) and Anxiety (Estimate = −.15, SE = .03, t = −4.77, p < .001) were obtained. Inspection of the estimates indicated that participants who were more anxious or avoidant reported being less happy at baseline.9 None of the three-way interactions were significant, contrary to our predictions. For example, anxious attachment did not predict greater changes in happiness following the compassionate exercise.
The model tested included the control variables (income, adherence, payment status, age) along with the exercise condition, followed by the attachment variables and interaction terms. Inspection of the random effects for this model revealed that the rate of change among participants tended to vary significantly (Estimate = .67, SE = .43, z = 1.58, p = .06) while controlling for all variables in the model. Initial status, or participants’ baselines values for depression did show significant variability (Estimate = 78.60, SE = 7.25, z = 10.85, p < .0001). However, the rate of change did not covary with baseline scores (Estimate = −1.05, SE = 1.52, z = −.69, p = .49), suggesting that participants’ depression at the outset of the study did not systematically impact their rate of change over time.
Main effects for Avoidance (Estimate = 3.13, SE = .57, t = 5.53, p < .001) and Anxiety (Estimate = 3.06, SE = .49, t = 6.28, p < .001) were obtained. Inspection of the estimates indicated that participants who were more anxious or avoidant were more depressed at baseline.10 None of the three-way interactions were significant, contrary to our predictions. That is, anxious participants did not respond differently to the compassion exercise in terms of changes in depression.
3.7.4 Multiple Regressions with Attachment
We had clear hypotheses for the relationships between attachment style, and response to exercise type. More specifically, we expected those with an anxious attachment style to particularly profit from the compassionate exercise. While the previous findings provide no evidence for this in the long run, we conducted regression models to determine possible effects in the short run. Three separate models were run involving well-being measures at 1 week (self-esteem, happiness and depression) regressed onto baseline values for each measure respectively, followed by the control variables, attachment style, condition, and the interaction term between attachment and condition.
Multilevel modeling provided evidence for the effectiveness of compassionate action in enhancing psychological well-being over time when compared to an inactive condition. Gains in happiness and self-esteem were found over the 6 month project for those in the active exercise group. They also marginally decreased in levels of depression. Over the 6 month period, the compassionate action intervention did not appear to be more effective for individuals with an anxious attachment style. However, at 1 week following the exercise period, those anxiously attached showed a greater reduction in depressive symptoms when in the compassionate action condition.
The primary aim of this study was to explore the potential benefits of other-directed compassionate actions for psychological well-being. Multilevel modeling revealed that participants who practiced the compassionate action exercise reported significantly greater increases in well-being over 6 months, as measured by levels of happiness and self-esteem, than participants who wrote about an early memory. These findings align well with previous studies demonstrating the positive influence of practicing compassion-rooted activities on mental health (e.g., Brown et al. 2003; Schwartz et al. 2003). Thus, our results buttress the use of brief, daily self-help exercises to increase well-being in a non-clinical population.
Participants in the compassionate action condition reported significantly greater increases in self-esteem at 6 months compared to those in the control condition. This finding lends support to previous research which found compassion-rooted activities to influence positive shifts in self-esteem (Krause and Shaw 2000; Raupp and Cohen 1992; Simmons 1991; Simmons et al. 1993). This, in turn, raises the following question: Why was the compassionate action exercise particularly effective in elevating self-esteem? According to the sociometer theory of self-esteem—which likens self-esteem to an internal measure of one’s relational value and success (Leary et al. 1995)—the current exercise possibly improved self-esteem by creating the conditions for achieving greater social appeal. Consistent with previous studies demonstrating the causal relationship between compassionate behaviours and charitable self-evaluations (e.g., O’Malley and Andrews 1983; Otake et al. 2006), adopting a compassionate social role may have also resulted in a favorable view of self as encompassing culturally valued qualities.
4.2 Subjective Happiness
The compassionate action condition led to significantly greater increases in happiness up to 6 months, thus indicating the enduring benefits of the exercise on positive mood states. This is an important finding, as happiness has been shown to engender a variety of positive life outcomes, including improved mental and physical health, work life, and interpersonal relationships (see Lyubomirsky et al. 2005 for a review). Furthermore, our findings are consistent with studies that found helping behaviours to predict positive emotions and mood states (e.g., Dulin and Dominy 2008; Millar et al. 1988; Williamson and Clark 1989; Yinon and Landau 1987); including happiness (e.g., Thoits and Hewitt 2001). Additionally, the differential effects of the compassionate action exercise on levels of happiness are reminiscent of previous findings that compassion-based behaviours have an upward influence on positive but not negative affect (e.g., Dulin and Dominy 2008).
The compassionate action exercise might have effected positive changes to one’s happiness levels by satisfying some of the defining components of happiness as outlined by Seligman (2002)—’the pleasant life’ and ‘the meaningful life’ (Seligman et al. 2006). For example, by representing a positive and valuable presence in others’ lives may have led individuals to happiness by conferring a sense of meaning and purpose. This finds support in research indicating that compassionate activities increase meaning (e.g., Greenfield and Marks 2004), which in turn elevate levels of happiness (e.g., Compton 2000). Given that happiness is conceptualized as being composed of three distinct types (Seligman 2002), using a measure such as the Orientations to Happiness measure (Peterson et al. 2005) that is designed to yield three separate subscale scores for Pleasure, Meaning, and Engagement, would yield important, detailed information regarding the relationship between compassionate action and happiness.
4.3 Depressive Symptoms
Participants in the compassionate action condition reported greater reductions in depressive symptomatology over time, but the difference with the control condition did not reach significance. On the one hand, this outcome is divergent from positive findings demonstrating a causal relation between lowered depressive symptoms and such expressions of compassion as volunteerism (e.g., Musick and Wilson 2003). Future empirical work needs to establish the type of self-help interventions that would be particularly helpful in reducing depressive symptoms in the general population.
4.4 The Influence of Attachment Style on Outcome
The interaction between exercise type and attachment style were tested in multilevel models. There was no evidence for the superiority of the compassionate goal condition over the control group for those high on anxious or avoidant attachment. More specifically, anxiously attached individuals did not show lasting improvements when in the compassion condition. However, regression models indicated that after the 1-week experimental period, anxious persons reported greater reductions in their depressive symptoms when in the compassion exercise. Future research is required to determined whether the practice of compassion is beneficial for anxiously attached individuals.
While this study presents several promising findings, a number of limitations must be acknowledged. Firstly, the compassionate action condition was compared to an inactive control group. While this allows us to conclude with greater confidence that gains in well-being evidenced by participants were not simply the result of placebo effects or demand characteristics, it is not possible to ascertain whether such gains were actually due to compassionate actions. For example, participants in the experimental condition had to exert greater effort than the control group by being actively engaged in new behaviours every day. In order to avoid dissonance, these participants may have reported higher levels of well-being to justify their efforts. Furthermore, the practice of a new behavior may have strengthened the ‘self-regulatory muscle’ (Muraven et al. 1999) and the practice of self-control may have created an upward effect on mood, such as increases in happiness and self-esteem (Tice et al. 2007). In short, participants in the experimental condition may have improved as a result of a number of factors unrelated to enacting compassionate behaviors.
Another limitation of this study is the high rate of attrition which appears to plague longitudinal internet-based research (e.g., Richards and Alvarenga 2002; Strom et al. 2000). This could have been due to a number of technical issues associated with the use of the internet, including changes to one’s daily internet access or one’s email address. Lastly, asking participants to repeat the same activities every night may have been perceived as a burden, or led participants to a point of monotony and loss of interest.
Finally, this study cannot escape the criticism that our results are of limited generalizability. Firstly, the sample was restricted to self-selected individuals that were largely Canadian, Caucasian, female, and highly educated, thus limiting the extension of our findings to the overall population. Furthermore, the entire study—from recruitment to assessment to intervention delivery—was conducted online. Thus, the results of this study cannot be generalized to individuals without daily internet access or knowledge in the use of computers.
To conclude, a 1-week, online exercise involving practicing other-directed compassion was found to provide lasting increases in happiness and self-esteem in a non-clinical population. These findings are harmonious with positive psychology objectives that focus on the cultivation of human strengths (Seligman et al. 2005), and highlight the possibility of incorporating the provision of compassion as a method of increasing emotional well-being.
Those under 18 years of age were not invited to participate in the study. However two participants who were 17 years old were included in the sample.
We investigated the possibility that participants’ mood at baseline (i.e. happiness, self-esteem, and depression) might predict completion or non-completion based on the exercise assigned. Several regression models were run predicting completion at 1 week, 1, 3 and 6 months with mood (SHI, RSES, CESD) and exercise condition as predictors. None of the interaction between participants’ mood and the condition assigned predicted completion of the project.
The relationship between baseline affect and degrees of freedom varied slightly across t tests because some participants failed to provide some demographic information.
The multilevel model in the prediction of self-esteem produced a main effect for income (Estimate = .07, SE = .02, t = 4.43, p < .001), and for payment status (Estimate = .23, SE = .08, t = 2.94, p = .003). Those with a higher income and those who were paid reported higher levels of self-esteem at baseline.
The multilevel model in the prediction of happiness produced a main effect for income (Estimate = .09, SE = .02, t = 4.79, p < .001), and for payment status (Estimate = .26, SE = .09, t = 2.74, p = .006). Those with a higher income, and those who were paid reported higher levels of happiness at baseline. However, a Time by payment status effect was also obtained (Estimate = −.03, SE = .01, t = −1.97, p = .05) and inspection of the estimates indicated those who were paid did not increase in happiness as much as those who joined the study without payment.
The dependent variables in this study (happiness, self-esteem and depression) were highly correlated, raising the question of independence between the constructs. It is possible, for example, that the increase in both happiness and self-esteem represent redundant effects. This issue was addressed in HLM by using “time-varying” predictors, which allows for the prediction of change in one outcome variable while controlling for co-occurring changes in the other “time-varying” predictor. In the first model, happiness was treated as the time-varying predictor of change in self-esteem. The results indicated that fluctuations in happiness significantly accounted for increases in self-esteem (Estimate = .07, SE = .03, t = 2.10, p = .04). The second model treated self-esteem as the time-varying predictor of change in happiness. The results indicated that fluctuations in self-esteem did not significantly account for increases in happiness (Estimate = .05, E = .04, t = 1.55, p = .12). These results suggests that happiness and self-esteem are inter-related but not equivalent.
The multilevel model in the prediction of depression produced a main effect for income (Estimate = −1.42, SE = .29, t = −4.91, p < .001), and for payment status (Estimate = −4.49, SE = 1.51, t = −2.98, p = .003). At the outset, those with a higher income, and those who were paid for participation reported lower levels of depression. Furthermore, a main effect for condition was obtained (Estimate = −7.21, SE = 2.29, t = −3.15, p = .002), indicating that those in the compassion group started the study feeling less depressed. The random effects for this model indicated that participants’ baseline depression scores were unrelated to their trajectory over time. Therefore, group differences at baseline should not have biased the rate of change rate obtained for the compassion and control group.
The longer multilevel model in the prediction of self-esteem also produced a main effect for income (Estimate = .06, SE = .01, t = 4.38, p < .001) such that wealthier participants had higher levels of self-esteem at baseline.
The longer multilevel model in the prediction of happiness produced a main effect for income (Estimate = .08, SE = .02, t = 4.47, p < .001), and for age (Estimate = −.009, SE = .004, t = −2.43, p = .02). Those who were younger and with a higher income, reported higher levels of happiness at baseline. A Time by payment status effect was also obtained (Estimate = −.03, SE = .01, t = −2.09, p = .04) and inspection of the estimates indicated those who were paid did not increase in happiness as much as those who joined the study without any payment.
The longer multilevel model in the prediction of depression produced a main effect for income (Estimate = −1.13, SE = .26, t = −4.28, p < .001), and a Time by age interaction effect (Estimate = .03, SE = .01, t = 2.10, p = .04). Younger participants reported greater reductions in depressive symptoms over time than older participants.
The main effects obtained in the regression analyses are not repeated here as they were previously reported in the multilevel modeling analyses.
This project was funded by a grant to the first author from the Social Sciences and Humanities Research Council of Canada. The article is based on portions of the second author’s honours’ thesis which was supervised by the first author.
- Allen, N. B., & Knight, W. E. J. (2005). Mindfulness, compassion for self, and compassion for others: Implications for understanding the psychopathology and treatment of depression. In P. Gilbert (Ed.), Compassion: Conceptualisations, research and use in psychotherapy (pp. 121–147). London: Routledge.Google Scholar
- Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and anger. New York: Basic Books.Google Scholar
- Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult romantic attachment: An integrative overview. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 46–76). New York: Guilford Press.Google Scholar
- Carmines, E. G., & Zeller, R. A. (1979). Reliability and validity assessment. Beverly Hills: Sage.Google Scholar
- Compton, W. C. (2000). Meaningfulness as a mediator of subjective well-being. Psychological Reports, 87, 156–160.Google Scholar
- Corcoran, K., & Fisher, J. (1987). Measures for clinical practice: A source book. New York: Free Press.Google Scholar
- Dulin, P. L., & Dominy, J. B. (2008). The influence of feeling positive about helping among dementia caregivers in New Zealand: Helping attitudes predict happiness. Dementia: The International Journal of Social Research and Practice, 7, 55–69.Google Scholar
- Gillath, O., Shaver, P. R., & Mikulincer, M. (2005). An attachment-theoretical approach to compassion and altruism. In P. Gilbert (Ed.), Compassion: Conceptualisations, research and use in psychotherapy (pp. 121–147). London: Routledge.Google Scholar
- Krause, N. M., Herzog, A. R., & Baker, E. (1992). Providing support to others and well-being in later life. Journals of Gerontology, 47, 300–311.Google Scholar
- Lyubomirsky, S., Sheldon, K. M., & Schkade, D. (2005b). Pursuing happiness: The architecture of sustainable change. Review of General Psychology. Special Issue: Positive Psychology, 9, 111–131.Google Scholar
- Mikulincer, M., & Shaver, P. R. (2007). Attachment in adulthood: Structure dynamics, and change. New York: Guilford Press.Google Scholar
- Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.Google Scholar
- Seligman, M. E. P. (2002). Authentic happiness: Using the new positive psychology to realize your potential for lasting fulfillment. New York: Free Press.Google Scholar
- Wang, S. (2005). A conceptual framework for integrating research related to the physiology of compassion and the wisdom of Buddhist teachings. In P. Gilbert (Ed.), Compassion: Conceptualisations, research and use in psychotherapy (pp. 121–147). London: Routledge.Google Scholar
- Yarcheski, A., & Mahon, N. E. (1989). A causal model of positive health practices: The relationship between approach and replication. Nursing Research, 38, 88–93.Google Scholar
- Yogev, A., & Ronen, R. (1982). Cross-age tutoring: Effects on tutors’ attributes. Journal of Educational Research, 75, 261–268.Google Scholar