Available online at www.sciencedirect.

com

ScienceDirect
Behavior Therapy 48 (2017) 778 – 792
www.elsevier.com/locate/bt

A Meta-Analysis of Compassion-Based Interventions: Current State

of Knowledge and Future Directions
James N. Kirby
Cassandra L. Tellegen
Stanley R. Steindl
The University of Queensland

examination of compassion across the lifespan. Although

Objective: Scientific research into compassion has bur- further research is warranted, the current state of evidence
geoned over the past 20 years and interventions aiming to highlights the potential benefits of compassion-based
cultivate compassion towards self and others have been interventions on a range of outcomes.
developed. This meta-analysis examined the effects of
compassion-based interventions on a range of outcome
measures. Method: Twenty-one randomized controlled Keywords: compassion; self-compassion; intervention; RCT;
meta-analysis
trials (RCTs) from the last 12 years were included in the
meta-analysis, with data from 1,285 participants analyzed.
Effect sizes were standardized mean differences calculated COMPASSION IS NOT A NEW CONCEPT; it has been
using the difference in pre-post change in the treatment discussed for thousands of years by ancient spiritual
group and control group means, divided by the pooled and religious traditions (Goetz, Keltner, & Simon-
pre-intervention standard deviation. Results: Significant Thomas, 2010; Kirby & Gilbert, 2017). What is
between-group differences in change scores were found becoming increasingly noticeable is the attention
on self-report measures of compassion (d = 0.55, k = 4, compassion is receiving by the scientific community
95% CI [0.33-0.78]), self-compassion (d = 0.70, k = 13, (Gilbert, 2014; Singer & Bolz, 2013). Over the
95% CI [0.59-0.87]), mindfulness (d = 0.54, k = 6, 95% last 20 years, research has shown a number of
CI [0.38-0.71]), depression (d = 0.64, k = 9, 95% benefits and positive associations of compassion
CI [0.45-0.82]), anxiety (d = 0.49, k = 9, 95% CI for our physiological health, including influencing
[0.30-0.68]), psychological distress (d = 0.47, k = 14, genetic expression in cross-sectional studies (e.g.,
95% CI [0.19-0.56]), and well-being (d = 0.51, k = 8, 95% Fredrickson et al., 2013), as well as in intervention
CI [0.30-0.63]). These results remained when including studies (Klimecki, Leiberg, Ricard, & Singer, 2014),
active control comparisons. Evaluations of risk of bias positive correlations found for mental health and
across studies pointed towards a relative lack of publication emotion regulation (e.g., MacBeth & Gumley, 2012),
bias and robustness of findings. However, the evidence as well as intervention studies showing benefits
base underpinning compassion interventions relies predom- (e.g., Jazaieri et al., 2013; Seppala, Rossomando,
inantly on small sample sizes. Conclusions: Future direc- & Doty, 2013), and associations between positive
tions are provided for compassion research, including the interpersonal and social relationships (e.g., Yarnell
need for improved methodological rigor, larger scale RCTs, & Neff, 2013), as well as longitudinal studies
increased specificity on the targets of compassion, and showing improvements (e.g., Crocker & Canevello,
2012). In light of significant positive associations
Address correspondence to James N. Kirby, Ph.D., School of and benefits associated with compassion, a number
Psychology, The University of Queensland, St Lucia, Brisbane, of compassion-based interventions have been devel-
4072; e-mail: [email protected]. oped that specifically aim to cultivate compassion
0005-7894/© 2017 Association for Behavioral and Cognitive Therapies. (e.g., Gilbert, 2014; Neff & Germer, 2013; Jazaieri
Published by Elsevier Ltd. All rights reserved. et al., 2013).
 compassion meta-analysis 779

defining compassion cultivation of compassion (Kirby, 2016): Compas-

Definitions of compassion vary, with some defining sion Focused Therapy (CFT; Gilbert, 2014), Mind-
it as an emotion (Goetz et al., 2010), others as a ful Self-Compassion (MSC; Neff & Germer, 2013),
multidimensional construct (Jazaieri et al., 2013; Compassion Cultivation Training (CCT; Center for
Strauss et al., 2016), and others as a motivational Compassion and Altruism Research and Education,
system (Gilbert, 2014). Goetz and colleagues (2010) 2015); Cognitively-Based Compassion Training
specifically define compassion as “the feeling that (CBCT; Pace et al., 2009), Cultivating Emotional
arises in witnessing another’s suffering and that Balance (CEB; Kemeny et al., 2012), and Loving-
motivates a subsequent desire to help” (p. 351). This Kindness (LKM) and Compassion Meditations
definition emphasizes compassion as an emotion; (CM; e.g., Wallmark, Safarzadeh, Daukantaite, &
however, among emotion scientists, only 20% agree Maddux, 2013). We have included in Appendix A
that compassion is an emotion, compared to over (see supplementary materials) a description of the
80% agreement on other emotions such as anger, elements included in each of these intervention
fear, disgust, sadness (Ekman, 2016). Indeed, Geshe approaches. Although all these interventions are
Thupten Jinpa, who developed the Stanford Com- secular in their design, theoretically these interven-
passion Cultivation Training program, defines tions have been typically influenced by Tibetan
compassion as being a complex multidimensional Buddhist traditions and perspectives of human
construct that is comprised of four key components: suffering (Hangartner, 2013). CFT is notably dif-
(a) an awareness of suffering (cognitive component), ferent from the other interventions, as the theoret-
(b) sympathetic concern related to being emotion- ical underpinning also includes a combination of
ally moved by suffering (affective component), (c) a evolutionary psychology, attachment theory, and
wish to see the relief of that suffering (intentional social mentality theory (Gilbert, 2014; Kirby, Doty,
component), and (d) a responsiveness or readiness to Petrocchi, & Gilbert, 2017). To date, all six forms
help relieve that suffering (motivational component; of interventions have been subject to the gold-
Jazaieri et al., 2013). Paul Gilbert, who developed standard evaluations of randomized controlled
Compassion-Focused Therapy, defines compassion trials (RCTs). Despite the increasing interest and
as “the sensitivity to suffering in self and others use of compassion-based interventions, particularly
(engagement), with a commitment to try to alleviate over the last 5 to 10 years when many of the RCTs
and prevent it (action)” (Gilbert, 2014, p. 19). In a have been conducted (Kirby, 2016; Leaviss & Uttley,
recent review, Strauss and colleagues (2016) sug- 2015), it remains unknown whether the evidence
gested that compassion includes five elements: base underpinning compassion-based interventions
(a) recognizing suffering; (b) understanding the demonstrates reductions in suffering and improve-
universality of suffering in human experience; ments in mental health.
(c) feeling empathy for the person suffering and
connecting with the distress (emotional resonance); aim
(d) tolerating uncomfortable feelings aroused in Despite compassion-based interventions being in-
response to the suffering person (e.g., distress, creasingly used by practitioners to help with the
anger, fear); and (e) motivation to act/acting to cultivation of compassion and improvement of
alleviate suffering. The notion of self-compassion well-being, there has been no synthesis of the data
has received increasing attention with the work of to date. The objective of this meta-analysis is to
Kristen Neff, who defined self-compassion, based on synthesize for the very first time the impacts of
her interpretations of Buddhist teachings, as having all compassion-based interventions in order to best
three components: (a) being mindful, rather than understand their overall effectiveness. Studies in-
overidentifying with problems; (b) connecting with cluded in this review were RCTs, involving adults
others, rather than isolating oneself; and (c) adopting wherein cultivating compassion towards self or
an attitude of self-kindness, rather than being others was a key component in the intervention.
judgmental (Neff, 2003). Given the differing defini- There were two major aims. The first aim was to
tions of compassion, it is not surprising that several evaluate the success of compassion-based inter-
different interventions have been developed to help ventions using meta-analytic techniques on the
cultivate compassion for self and others. following seven outcome variables: (1) compassion,
(2) self-compassion, (3) mindfulness, (4) depression,
current compassion-based (5) anxiety, (6) psychological distress, and (7) well-
interventions being. The second aim was to conduct moderator
A recent critique of compassion-based interventions analyses to examine impact of potential variables
identified that there are at least six currently em- on outcomes, including gender, age, intervention
pirically supported interventions that focus on the length, involvement of program developer, country
780 kirby et al.

paper was published, severity of population being PsycINFO, PsycARTICLES, PsycBOOKS, PubMed,
examined, and attrition. However, it should be noted ERIC and ProQuest Dissertations and Thesis Global.
that this aim was not able to be achieved due to a lack The search was conducted on January 18, 2017.
of available data. In an attempt to collect all eligible papers, we also
emailed on multiple occasions active compassion
Method intervention researchers from MSC, CEB, CCT,
protocol and registration CBCT, and CFT for unpublished papers or disserta-
The review protocol was prospectively registered tions. English was the language selected and the
in PROSPERO (Kirby, Tellegen, & Steindl, 2015), following fields were searched for in any field:
and our meta-analysis followed the standards of “compassion” AND “program,” “therapy,” “inter-
the PRISMA guidelines (Moher, Liberati, Tetzlaff, vention,” “training.” Studies were screened by the
& Altman, 2009). There was no funding for this first author based on title/abstract. Abstracts and
meta-analysis. full-text articles were then examined by the first and
second authors to determine if studies met inclusion
eligibility criteria criteria. Any uncertainties regarding eligibility for
To be included in the meta-analysis, studies had to inclusion were resolved by discussion between the
meet the following eligibility criteria: (a) the study first, second, and third authors.
evaluated an intervention where one of the explicit
main objectives of the intervention was to purposively data extraction
generate compassion or self-compassion; (b) had to The first and second authors extracted data and
be greater than one stand-alone session (e.g., multiple study characteristics. Both authors extracted data
sessions or one session followed by further compas- independently, with any discrepancies resolved by
sion exercises or homework); (c) written in English; discussion. The following information on study
(d) published in a peer-reviewed journal or in a characteristics was extracted: year published, groups
dissertation; (e) included adult participants only; included in RCT, delivery format, intervention
(f) the intervention was evaluated in an RCT; and description and length, intervention completion
(g) included at least one self-report measure related data, sample criteria, measurement time points,
to the outcomes of compassion, self-compassion, sample size, participant age information, percentage
mindfulness, depression, anxiety, psychological dis- of males, developer involvement (a program devel-
tress, or well-being. Studies without outcomes related oper was an author of the paper vs. not), country
to these specified constructs (e.g., only reporting on from which participants were recruited, attrition
neuroimaging data) were excluded. rates at postintervention, protocol adherence data,
For the study to be included in the meta-analyses, and outcome measures. For the meta-analyses,
the paper needed to report sufficient data for stan- the following data were extracted: means, standard
dardized mean difference effect size calculations deviations, and sample sizes for each group at pre-
(i.e., means, standard deviations, and sample sizes and postintervention.
for each group at preintervention and postinterven-
tion; when this could not be extracted we contacted quantitative analyses
authors for the data) and needed to report on data A series of analyses were performed combining effect
comparing the intervention to a control group sizes calculated across seven outcome categories.
(e.g., waitlist control or active control). We calculated
the performance of compassion-based interventions Outcome Categories
compared to waitlist control conditions (i.e., involv- The dependent variables in the studies were
ing no intervention). We then also calculated how classified into seven different outcome categories:
compassion-based interventions compared when (1) compassion; (2) self-compassion; (3) mindful-
including studies that had an active control group ness; (4) depression; (5) anxiety; (6) psychological
(i.e., those involving some form of intervention). distress; and (7) well-being. Analyses were con-
As active controls would be likely to result in ducted separately for each outcome category. The
improvements themselves, we hypothesize that this various measures included within each outcome
will produce smaller effect sizes than if compared to category are detailed in Appendix B.
waitlist controls (e.g., Cuijpers, Cristea, Karyotaki, Effect Size Calculations
Reijnders, & Huibers, 2016; Khoury et al., 2013). The effect sizes were standardized mean differences,
represented by d, and can be interpreted using
search strategy Cohen’s (1992) guidelines of small (0.2), medium
Several strategies were employed to obtain relevant (0.5), and large (0.8) effects. Effect sizes were calcu-
studies. First, the following databases were searched: lated based on the pre-post change in the treatment
 compassion meta-analysis 781

group means minus the pre-post change in the was not conducted because we were not able to
control group means, divided by the pooled pre- obtain accurate estimates of the population corre-
intervention standard deviation (Carlson & Schmidt, lations between categories to compute covariances
1999; Morris, 2008). This approach, which com- between effect sizes (Cheung, 2013; Gleser &
pares changes across groups from pre- to post- Olkin, 2007). Computation of overall effect sizes
intervention, was chosen as it includes all the was based on a weighted-average of the effect sizes
information available in the study as opposed to using a random-effects model. The random-effects
comparing group means at postintervention. This model was chosen as it assumes that variation
approach was selected as research has found this between studies can be systematic and not only due
calculation to provide the best estimate in terms of to random error (Borenstein, Hedges, Higgins, &
bias, precision, and robustness to heterogeneity of Rothstein, 2009). This assumption fits with the
variance (Morris, 2008). This approach also gives data in this study as it is likely that the true effect
increased precision on estimates of treatment effects of interventions will vary depending on character-
and is able to statistically account for any preinter- istics of the sample and implementation of the
vention differences between groups (Morris, 2008). intervention.
The pooled preintervention standard deviation was To examine if there was significant variation of
chosen as the denominator in the formula, as using effect sizes between studies, the Q-test for hetero-
it has been shown to provide an unbiased estimate geneity was computed (Hedges & Olkin, 1985) and
of the population effect size and has a known evaluated against a chi-squared distribution with
sampling variance (Morris, 2008). See Appendix C df = k – 1 (where k = number of studies). A sig-
for formulae. nificant Q statistic indicates significant variability
among effect sizes. As the Q statistic is dependent
Multiple Effect Sizes per Study
on the number of studies, the I 2 index was also
Some studies reported on multiple measures within
computed to provide a measure of the degree of
the same outcome category (e.g., two measures
heterogeneity. I 2 is interpreted as the percentage
of psychological distress). It is recommended that
of variability among effect sizes across studies due
only one effect size per study is included in a
to heterogeneity rather than chance/sampling error.
meta-analysis, otherwise each data point will not
The I 2 index can be interpreted as follows: 0%
be independent (Borenstein et al., 2009). As such,
indicates homogeneity; 25% indicates small hetero-
when multiple scales are used for a construct in a
geneity; 50% is medium; and 75% is large (Huedo-
study, the effect sizes for these scales must be
Medina, Sanchez-Meca, Marin-Martinez, & Botella,
combined to create only one effect size for use in
2006).
the meta-analysis. The most accurate procedure
for combining multiple effect sizes from one study Moderator Analyses
requires estimates of the correlations between In the registered protocol for this systematic review,
dependent measures and such correlations have a it was specified that moderator analyses would
large impact on effect sizes generated (Bijmolt & examine gender, age, intervention length, involve-
Pieters, 2001; Marin-Martinez & Sanchez-Meca, ment of program developer, country paper was
1999). However, accurate estimates of correlations published, severity of population being examined,
between all pairs of scales were not obtainable. and attrition. However, at the completion of data
As such, a variance-weighted average of effect sizes collection we found that for all outcomes, except
from the scales within each study was used to for self-compassion and psychological distress, had
obtain one effect size for analysis. This procedure less than 10 studies contributing data. Furthermore,
is deemed acceptable when there is insufficient there was insufficient reporting on moderators
information to estimate correlations between de- such as dosage, attrition, and severity of initial
pendent measures and when the measures within problem, and there was a lack of variation among
each category are assumed to be highly correlated each moderator (for example, 13 of the 17 studies
and homogeneous indicators for the same outcome were group delivery format). As such, any moder-
(Marin-Martinez & Sanchez-Meca, 1999). ator analyses would involve comparisons of groups
with low numbers of studies and thus would lack
Analysis Strategy
sufficient power to detect moderation effects (Card,
The software used for the analyses was Microsoft
2012). Hence, it was not deemed appropriate to
Excel and Comprehensive Meta-Analysis (CMA;
conduct moderator analyses.
Borenstein, Hedges, Higgins, & Rothstein, 2005).
Meta-analytic statistics were conducted on the Risk of Bias Within Studies
seven outcome categories separately. A multivariate The PRISMA statement recommends that system-
meta-analysis looking at all outcomes concurrently atic reviews and meta-analyses include methods for
782 kirby et al.

assessing risk of bias within studies (Moher et al., sufficient data (n = 8). The remaining 21 papers
2009). The Cochrane risk of bias tool (Higgins were included in the quantitative meta-analysis,
et al., 2011) was used to evaluate the trials in the of these 4 papers included an active comparison
quantitative analyses, and was completed by the condition (n = 4), which we analyzed separately.
first and second authors independently and cross- Thus, we performed analysis of compassion-based
checked with a 95% agreement. This tool is neither interventions compared to waitlist control conditions
a scale nor a checklist. It is a domain-based (n = 17), and then with the inclusion of the active
evaluation, in which critical assessments are made comparison conditions (n = 21). See Figure 1 for
separately for each of seven different domains: the PRISMA flow diagram displaying the identifica-
sequence generation, allocation concealment, blind- tion and selection of studies for inclusion.
ing of participants and personnel, blinding of
outcome assessment, incomplete outcome data, quantitative results
selective outcome reporting, and “other issues.” Study Characteristics
Within each domain, what was reported to have Appendix D displays a summary of the study
happened in the study is described in sufficient characteristics. In the 21 papers included in the
detail to support a judgment about the risk of bias. quantitative analyses, 20 different trials were
This judgment can be “low risk” of bias, “high described (2 papers reported on different outcome
risk” of bias, or “unclear risk” of bias. data from the same trial; Jazaieri et al., 2013,
Jazaieri et al., 2014), with 21 different samples
Risk of Bias Across Studies
of participants evaluating an intervention (Kelly
Risk of bias across studies largely refers to the
et al., 2009, included two compassion intervention
possibility that null or negative results are less likely
groups vs. a waitlist control group). Of the 21
to be published, meaning that available data may be
samples, 5 interventions were based on MSC,
biased (e.g., publication bias and selective reporting
6 were based on LKM and CM, 3 based on CFT,
bias; Liberati et al., 2009). Three steps were taken
2 programs informed by MBCT (referred to as
to evaluate risk of bias across studies. First, funnel
Compassion-Mindfulness Therapy; C-MT), 2 in-
plots with the effect size plotted against the inverse
terventions were based on CCT, 2 interventions
of the standard errors were inspected to determine
were self-directed self-compassion interventions
if there was selective reporting of small studies with
with writing modules informed by Neff’s con-
larger effect sizes. Second, trim and fill analyses
ceptualization of self-compassion, and 1 interven-
were conducted by imputing values in the funnel
tion was based on Cognitively-Based Compassion
plot to make it symmetrical and computing a
Training. Fifteen of the interventions used a group
corrected effect size estimate (Duval & Tweedie,
delivery format and 6 interventions were self-
2000). Third, Orwin’s (1983) failsafe N was calcu-
directed. The minimum hours of intervention
lated to determine the number of studies with
dosage for each of the 16 interventions for which
null results (set at d = 0) needed to reduce the effect
it could be determined ranged from 1 to 20 hours.
size to the smallest meaningful effect size (chosen
There were 4 studies with an active control
as d = 0.10).
condition, which included 3 that were mindfulness
Results meditation interventions, and 1 that was a coping
study selection with stress intervention.
The searches yielded a total of 1,918 papers. After All studies were conducted within a 12-year
removing duplicates, 1,510 papers remained. After period (2005–2017). A total of 1,285 participants
assessing for eligibility, 1,439 studies were excluded were included in the trials with sample sizes
as they were not an intervention study (n = 1,236), ranging from 24 to 228. Nine trials had partici-
not in English (n = 12), not an adult population pants from the USA, 2 trials were from Canada,
(n = 5), or did not include a compassion-based and 1 trial with participants from each of Chile,
intervention (n = 186). After screening, 71 full-text Greece, Hong Kong, Israel, Japan, Korea, the
articles were assessed for eligibility and 42 were Netherlands, New Zealand, Spain, and Sweden.
excluded as they were not an RCT (n = 34), did not Five of the trials had only female participants and
have full-text available (n = 4), did not include any there was a mean of 26% of males across trials
compassion or well-being outcome (n = 2), or only (SD = 23.16, range = 0 to 100%). Across 20 trials,
evaluated a single-session lab-based experiment the mean participant ages ranged from 18.8 to
without homework exercises (n = 2). Twenty-nine 51.1 years, with an average of 34.91 years (SD =
papers were assessed for quantitative data so that 10.47). Across the 20 trials with data, there was
it could be meta-analyzed, and 8 of these papers an average of 17.5% attrition in the treatment
could not be included, as the paper did not report group from pre- to postintervention (range = 0 to
 compassion meta-analysis 783

FIGURE 1 PRISMA flow chart describing identification and selection of studies for inclusion in the meta-analysis adapted from
Moher, Liberati, Tetzlaff and Altman (2009).

45.6%). Ten of the 20 trials reported on follow-up adherence. High rates of protocol adherence were
data, with follow-up periods ranging from 1 to reported in the 4 trials where it was available.
12 months postintervention (note that follow-up
data were not analyzed in this review). Twelve compassion-based interventions
trials included a program developer as an author compared to waitlist
and 8 trials did not have a program developer as control conditions
an author. Protocol adherence data (examining Between-Group Differences in Change Scores
therapist delivery of intervention) were not rele- Table 1 summarizes the effect sizes for each
vant for the 6 self-directed interventions and 8 of sample for each outcome. Table 2 displays the
the remaining 15 trials did not report on protocol effect sizes for the compassion-based interventions
784 kirby et al.

Table 1
Standardized Mean Difference Effect Sizes and Sample Sizes for Each Study Sample Per Outcome Category Using Waitlist
Control Conditions
Study sample Effect size (d)
nT nC Compassion Self-compassion Mindfulness Depression Anxiety Psychological Well-being
Distress
Albertson et al. (2014) 98 130 - 0.46 - - - - -
Arimitsu (2016) 19 16 - 0.99 - - - 0.52 0.51
Carson et al. (2005) 18 25 - - - - - 0.05 -
Friis et al. (2016) 32 31 - 0.57 - 0.45 - 0.99 -
Jazaieri et al. (2013) 50 30 0.48 0.54 - - - - -
Jazaieri et al. (2014) 50 30 - - 0.44 - 0.49 0.02 0.42
Kelly et al. (2009) 23 24 - - - 0.14 - 0.51 -
(Self-soothing)
Kelly et al. (2009) 26 24 - - - 0.49 - 0.53 -
(Attack-resisting)
Lee & Bang (2010) 30 30 - 0.88 0.78 0.72 0.90 0.93 1.17
Lo et al. (2013) 41 41 - - - 0.94 0.75 0.59 0.66
Mosewich et al. (2013) 29 22 - 0.74 - - - 0.64 -
Neff & Germer (2013) 24 27 0.63 1.34 0.52 0.91 0.74 0.39 0.31
Pons (2014) 26 24 0.53 1.22 0.76 0.79 0.49 0.96 0.38
Shahar et al. (2015) 14 18 - 0.58 - 0.57 0.02 0.37 0.55
Smeets et al. (2014) 27 25 - 1.04 0.79 - 0.10 0.28 0.26
Toole & Craighead (2016) 40 40 - 0.17 - - - - -
Wallmark et al. (2013) 20 22 - 0.74 0.44 - - 0.52 -
Weibel (2008) 33 25 0.71 0.38 - - 0.27 - -
Note. d = standardized mean difference effect size; nT = sample size of treatment group; nC = sample size of control group; cells left blank
when outcome not assessed by study

on each outcome category. All analyses were anxiety, d = 0.49, k = 9, 95% CI [0.30-0.69],
conducted using a random effects model. Overall p b .001, and for psychological distress, d = 0.47,
significant moderate effects were found for all k = 14, 95% CI [0.19-0.56], p b .001. Finally,
outcome categories. A significant moderate effect a significant moderate effect size was found for
size was found for the compassion outcome well-being, d = 0.51, k = 8, 95% CI [0.30-0.63],
category, d = 0.55, k = 4, 95% CI [0.33-0.78], p = .001.
p b .001, and the self-compassion outcome There was a significant amount of heterogeneity
category, d = 0.70, k = 13, 95% CI [0.53-0.87], with a medium percentage of variability in effect
p b .001. A significant moderate effect size was sizes for self-compassion, Q(12) = 30.00, p = .003,
also found for mindfulness, d = 0.54, k = 6, 95% I 2 = 59.99. Tests for heterogeneity were not sig-
CI [0.38-0.71], p b .001. Similarly, a significant nificant for the remaining six outcomes and only
moderate effect size was found for depression, null-to-small percentages of variability in effect sizes
d = 0.64, k = 9, 95% CI [0.45-0.82], p b .001, were detected.

Table 2
The Effects of Compassion-Based Interventions on Outcome Categories Using Waitlist Control Conditions
Outcome category k N d (overall effect size) d Lower 95% CI d Upper 95% CI z p (for d) Q p (for Q) I2
Compassion 4 239 0.55⁎⁎⁎ 0.33 0.78 4.828 N .001 0.61 0.893 0.00
Self-Compassion 13 882 0.70⁎⁎⁎ 0.53 0.87 8.007 N .001 30.00⁎⁎ 0.003 59.99
Mindfulness 6 335 0.54⁎⁎⁎ 0.38 0.71 6.561 N .001 3.38 0.642 0.00
Depression 9 470 0.64⁎⁎⁎ 0.45 0.82 6.807 N .001 6.83 0.556 0.00
Anxiety 9 500 0.49⁎⁎⁎ 0.30 0.68 4.987 N .001 9.22 0.324 13.24
Psychological Distress 14 738 0.47⁎⁎⁎ 0.19 0.56 3.907 N .001 14.38 0.109 37.40
Well-being 8 442 0.51⁎⁎ 0.30 0.63 5.503 N .001 25.63⁎ 0.019 49.28
Note. d = standardized mean difference effect size; Q = test statistic for heterogeneity; k = number of samples; N = participants contributing
to outcome; p = test for significance evaluated against .05; I2 = measure of degree of heterogeneity; z = z-score.
* p b .05, ** p b .01, *** p b .001
 compassion meta-analysis 785

Risk of Bias Within Studies and psychological distress outcomes. There was a
The results of the evaluation for risk of bias within slight trend for less precise studies with smaller
studies are displayed in Figure 2. All studies were sample sizes to be biased towards having larger
unable to blind participants to the intervention effect sizes. The trim and fill analysis for compas-
being received, indicating that performance bias sion suggested that two studies were missing and
might operate, a risk of bias common to psycho- computed a corrected effect size estimate (d = 0.49,
logical intervention research. The large majority of 95% CI [0.30-0.69]) slightly lower than that with-
studies did not report whether allocation to ran- out correction (d = 0.55, 95% CI [0.33-0.78]). The
domization was concealed or whether researchers trim and fill analysis for self-compassion imputed
were blind to outcome assessment. However, one five missing studies, computing a corrected effect
study reported a high risk of bias for blinding of size (d = 0.51, 95% CI [0.33-0.69]) lower than
outcome assessment and one study reported a low without correction (d = 0.70, 95% CI [0.53-0.87]).
risk of bias. For 10 out of 16 studies there was a low The trim and fill analysis for mindfulness imputed
risk of selection bias in terms of random sequence three missing studies finding a corrected effect
generation, with the remaining studies not report- size (d = 0.45, 95% CI [0.29-0.61]) slightly lower
ing how random sequencing was generated. than that without correction (d = 0.54, 95% CI
Reporting bias was unclear in all studies, with [0.38-0.71]). The trim and fill analysis for psycho-
the exception of one study that had a registered logical distress imputed five studies and found a
protocol and demonstrated no reporting bias. corrected effect size (d = 0.31, 95% CI [0.14-0.49])
Attrition bias was an unclear risk for most studies, lower than that without correction (d = 0.47,
with two studies categorized as low risk. A low risk 95% CI [0.30-0.63]). It is important to note that
of other sources of bias was identified across all all confidence intervals for the corrected effect size
trials. Overall, this evaluation points to insufficient estimates did not span zero, indicating significant
reporting in most papers regarding the majority effects with corrected effect sizes.
of risks of bias, with the exception of random Orwin’s failsafe N, indicating how many studies
sequence generation. with a null effect size would need to be located
to reduce the overall effect size to below d = 0.1,
Risk of Bias Across Studies was as follows for each outcome: compassion = 19,
Funnel plots showed no asymmetry for the depres- self-compassion = 62, mindfulness = 27, depression =
sion, anxiety, and well-being outcomes. Trim and 49, anxiety = 36, psychological distress = 42, and
fill analyses for these outcomes suggested that no well-being = 31. It is highly unlikely that such large
studies were missing and the effect size estimates numbers of studies with null results exist, indicating
remained unchanged. Some asymmetry was seen the robustness of the findings to publication bias.
for the compassion, self-compassion, mindfulness, Taken together, the evidence assessing risk of bias

FIGURE 2 Assessment of risk of bias within studies using the Cochrane risk of bias tool (Higgins et al., 2011).
786 kirby et al.

Table 3
Standardized Mean Difference Effect Sizes and Sample Sizes for Each Study Sample Per Outcome Category for Active Control
Conditions
Study sample Effect size (d)
nT1 nT2 Compassion Self-compassion Mindfulness Depression Anxiety Psychological Well-being
Distress
Desbordes et al. (2012) 12 12 - - - 0.30 -.45 - -
Feliu-Soler et al. (2016) 16 16 - 0.17 0.17 - - -.02 0.03
Held & Owens (2015) 13 14 - -.14 - - - -.26 -
Mantzios & Wilson (2014) 48 50 - 0.20 0.02 - - - -
Note. d = standardized mean difference effect size; nT = sample size of treatment group; nC = sample size of control group; cells left blank
when outcome not assessed by study

across studies suggests that the findings were not moderate effect size was found for well-being, d =
likely to be heavily influenced by publication bias. 0.48, k = 9, 95% CI [0.28-0.67], p = .001. There
was a significant amount of heterogeneity with a
compassion-based interventions medium percentage of variability in effect sizes for
with the inclusion of active self-compassion, Q(15) = 40.05, p = N .001, I 2 =
control conditions 62.55, and psychological distress, Q(15) = 34.94, p =
Table 3 summarizes the effect sizes for each sample 0.003, I 2 = 57.07. Tests for heterogeneity were not
for each outcome for the active control studies. significant for the remaining five outcomes and only
Table 4 displays the effect sizes for the compassion- null-to-small percentages of variability in effect sizes
based interventions on each outcome category were detected.
with the inclusion of the active control conditions.
All analyses were conducted using a random effects Discussion
model. No studies contributed to the compassion This is the first meta-analysis to investigate the
outcome category. Overall, there were slight de- effects of compassion-based interventions. Over-
creases in the effect sizes, however, all remained as all, meta-analytic techniques were performed on
significant moderate effect sizes. A significant mod- 21 RCT studies (containing data from 1,285
erate effect size was found for the self-compassion participants), which were conducted over a 12-year
outcome category, d = 0.60, k = 16, 95% CI period, from different countries around the world
[0.33-0.78], p b .001. A significant moderate effect (e.g., Canada, Chile, Greece, Hong Kong, Israel,
size was also found for mindfulness, d = 0.46, k = 8, Japan, Korea, Netherlands, New Zealand, Hong
95% CI [0.28-0.65], p b .001. Similarly, a significant Kong, Spain, Sweden, and United States). There were
moderate effect size was found for depression, d = significant pre-post intervention moderate effect
0.62, k = 10, 95% CI [0.44-0.80], p b .001, anxiety, sizes (standardized mean differences) for compas-
d = 0.42, k = 10, 95% CI [0.19-0.64], p b .001, sion, self-compassion, and mindfulness. Significant
and for psychological distress, d = 0.40, k = 16, moderate effects were also found for reducing
95% CI [0.23-0.57], p b .001. Finally, a significant suffering-based outcomes of depression, anxiety,

Table 4
The Effects of Compassion-Based Interventions on Outcome Categories When Including Active Control Conditions
Outcome category k N d d Lower d Upper z p (for d ) Q p (for Q) I2
(overall effect size) 95% CI 95% CI
Compassion 4 239 0.55⁎⁎⁎ 0.33 0.78 4.828 N .001 0.61 0.893 0.00
Self-Compassion 16 980 0.60⁎⁎⁎ 0.44 0.76 7.257 N .001 40.05⁎⁎ N .001*** 62.55
Mindfulness 8 465 0.46⁎⁎⁎ 0.28 0.65 4.923 N .001 10.12 0.182 30.80
Depression 10 506 0.62⁎⁎⁎ 0.44 0.80 6.799 N .001 7.57 0.578 0.00
Anxiety 10 536 0.42⁎⁎⁎ 0.19 0.64 3.608 N .001 14.97 0.092 39.89
Psychological Distress 16 797 0.40⁎⁎⁎ 0.23 0.57 4.615 N .001 34.94 0.003** 57.07
Well-being 9 474 0.48⁎⁎ 0.28 0.67 4.745 N .001 13.66⁎ 0.091 41.44
Note. d = standardized mean difference effect size; Q = test statistic for heterogeneity; k = number of samples; N = participants contributing
to outcome; p = test for significance evaluated against .05; I2 = measure of degree of heterogeneity; z = z-score.
* p b .05, ** p b .01, *** p b .001
 compassion meta-analysis 787

and psychological distress. Significant moderate Mosewich et al., 2013). In terms of the adapted
effects were also found for well-being. Risk of bias MBCT interventions (Lee & Bang, 2010; Lo et al.,
evaluations across papers using trim and fill analyses 2013), which had a specific compassion focus, it is
and Orwin’s failsafe N indicated that the findings unknown what specific mechanisms produced out-
were robust and not likely to be heavily influenced comes (for example, mindfulness- or compassion-
by publication bias. When including active control specific components). Thus, a clearer focus on the
comparisons, although the effect sizes slightly processes underpinning these interventions to deter-
decreased, they all remained as significant moderate mine the mechanisms of change are important to
effect sizes. Although the evidence-base underpin- understand the actual impact of the “compassion”
ning compassion-based interventions relies predom- elements within interventions.
inantly on small underpowered sample sizes, this is What is clear is that the demand for compassion-
a significant limitation. based interventions is increasing, with many
There are four key findings from this meta-analysis. clinicians and teachers being trained in various
First, the current evidence base for compassion-based compassion approaches (Kirby, 2016). However,
intervention is small. Despite identifying 71 evalua- the field of compassion science will be hampered
tion studies, only 29 were small-scale RCTs, of without greater consensus on how to define and
which only 4 studies used an active control condition. measure this construct. Although not the focus of
Second, the significant moderate effect sizes across this meta-analysis, our view is that an evolutionary
outcomes demonstrated the potential impact of model to understanding the emergence of compas-
compassion-based interventions; however, this find- sion offers one possible unifying framework for
ing is limited to largely nonclinical populations. the field of compassion science (for more on an
Third, there was great variability in the outcomes evolutionary approach, see Gilbert 1995). Another
measured, with few RCTs using compassion-based clear difficulty that stems from the differences in
self-report questionnaires. Fourth, there is a need for definition is how to most appropriately measure
greater methodological rigor and improved reporting compassion. Many of the interventions included in
in this intervention field. the meta-analysis did not measure compassion as
The development of outcome research for a new an outcome variable. In a recent review Strauss
intervention approach is time consuming (Sanders & et al. (2016) highlighted how there are a general
Kirby, 2014) and starts with initial feasibility studies lack of self-report measures available for measuring
(e.g., case studies, uncontrolled trials), moving to compassion, and this is one of the limitations of
small-scale RCTs comparing to a waitlist control or the field.
treatment-as-usual condition, and finally to large-
scale RCTs comparing to other effective treatments limitations
(e.g., CBT or ACT). Compassion-based interventions This review is limited by the number and quality of
are still within their infancy, only just commencing studies included. Employing the rigorous criteria of
small-scale RCTs. Nevertheless, the 21 RCT studies restricting the review to peer-reviewed published
produced overall standardized mean difference effect RCTs comes with the trade-off of excluding other
sizes for a range of important outcomes, indicating non-RCT evaluation studies, and 42 studies were
the results are at least promising. excluded due to this criterion. In future meta-
When considering these findings in the context analyses of compassion-based interventions open-
of the theoretical underpinnings of compassion, it ing to other databases, and also to clinical trials,
becomes clear that there is still a lack of clarity and registries may identify additional studies, as well as
agreed-upon processes on how to best define and help address possible issues related to publication
measure this construct. For example, the interven- bias. Search of clinical trials registry identified three
tions examined had varying definitions of compas- interventions that are currently in progress: CCT
sion and focused on different types of compassion for patients with chronic pain, MSC as a non-
(e.g., self or other), with MSC adopting a self- randomized effectiveness trial, and a randomized
compassion approach defined by Neff (2003), and trial of CFT for individuals with depression who
CCT using a multiconstruct definition of compassion identify as gay, lesbian, or bisexual.
defined by Jinpa (2015), focusing largely on guided A key aim was to examine the effects of the
meditations to cultivate compassion (Jazaieri et al., interventions on compassion outcomes; however,
2013). There were other interventions with no only four studies measured compassion as an
formal meditation/mindfulness components that outcome, limiting the generalizability and robust-
instead focused on psychoeducation, writing tasks ness of this outcome. Additionally, most studies
(e.g., letters), self-reflections, and imagery exercises relied on measures that assessed compassion or
to cultivate compassion (e.g., Kelly et al., 2009; self-compassion as a trait, not whether there was
788 kirby et al.

actually any change in motivation or action in assessment and allocation concealment). We list 12
compassion or self-compassion. As a result, the specific recommendations:
question remains as to whether compassion-based
interventions actually lead to greater compassion- 1. Use reliable and valid self-report question-
ate behaviors. naires of compassion or self-compassion as an
Our statistical analysis was restricted to self- outcome measure. In this review, only 20% of
report measures; thus, we could not include studies measured compassion as an outcome,
important studies that assessed compassion using whereas 76% measured self-compassion.
bodily measures or brain imaging (e.g., Desbordes 2. Use measures that have normative data with
et al., 2012; Mascaro, Rilling, Negi, & Raison, clinical cutoffs (e.g., Beck Depression Inventory;
2013) or studies that included behavioral measures BDI) in order to calculate clinical and reliable
(Leiberg, Klimecki, & Singer, 2011). An alternative change scores (e.g., Jacobson and Truax, 1991).
approach when meta-analyzing self-report data In our meta-analysis, 30% of the studies
where multiple measures exist for the same reported on the BDI, indicating that in the
construct (e.g., DASS and BDI for depression) is future it will be possible to evaluate this.
to use the measure with the strongest psychometric However, there are no current measures of
properties. We did not use this approach, as compassion or self-compassion with normative
research has suggested that when dealing with data—an important area of future research.
multiple effect sizes per study, procedures that use 3. Conduct RCTs with clinically diagnosed pop-
the complete set of measurements outperform those ulations (e.g., major depression, anxiety) to
that represent each study by a single value (Bijmolt determine the clinical utility of compassion-
& Pieters, 2001). A further limitation of this review based interventions. This could be achieved
was that we were unable to assess for the effect of using structured clinical interviews.
specific moderators (e.g., gender, dosage, interven- 4. Provide clear eligibility criteria guidelines for
tion delivery) or intervention components, largely RCT evaluations. When determining eligibility
due to the small number of studies per outcome. criteria for studies, assess for use of pharma-
In addition, we were not able to determine the cotherapy during both treatment and follow-
efficacy of compassion-based interventions for clin- up period. Also assess whether participants
ical populations. Our search was also restricted to have previous or ongoing formal meditation
English language, and we limited our criteria to experience/practice. Identifying religious back-
studies evaluating adults. The compassion interven- grounds would also be useful to determine
tion studies were conducted predominantly with whether this moderates impact or engagement
females, with only 26% of participants included in compassion interventions.
in the meta-analysis being male. The generalizability 5. Conduct RCTs that have adequately powered
of the findings to males is limited and more research sample sizes and that have the control condition
is warranted to examine the effects with males. not as a waitlist or treatment as usual, but with
an active comparison such as a mindfulness-
implications for compassion-based based intervention, Acceptance and Commit-
intervention research ment Therapy (ACT) or Cognitive Behavior
We will provide a series of recommendations for Therapy (CBT). When conducting such active
the field of compassion-based interventions in two comparisons, it is important to ensure that the
parts, the first in regards to methodological consid- protocols of the intervention are clearly de-
erations, the second in terms of future research. scribed to determine the differences between the
interventions and measurements are included
Methodological Considerations that examine the processes of change unique
This review highlights the need to improve the to the intervention. A recent RCT between
methodology and reporting within compassion- CBT and ACT by Craske and colleagues (2014)
based intervention research. In order that high- provides a useful example. Given potential
quality, adequately powered RCTs are conducted, overlaps between mindfulness and compassion-
important recommendations for compassion-based based interventions dismantling interventions
intervention research follow. The evaluation of risk or an appropriately designed process-focused
of bias within studies indicated some key areas investigation supplemented by adherence checks
where reporting of potential risks of bias were not and homework compliance would also be useful
clearly described in most studies. It is likely that to determine the differential outcomes.
some of the studies also did not follow the most 6. Researchers should follow (and report in the
rigorous RCT protocols (e.g., blinding of outcome published paper that they adhered to) the JARS
 compassion meta-analysis 789

(APA, 2008) or CONSORT (Moher et al., anxiety play a crucial role in determining if
2010) guidelines for RCTs. the individual has an anxiety disorder. Thus,
7. Use protocol adherence measures to deter- assessing for frequency and intensity of com-
mine the fidelity of intervention delivery. This passion in daily life may provide important
can be achieved by the therapist or practi- insights into understanding how compassion
tioner delivering the intervention recording is experienced, and how differing levels of
their fidelity in alignment with the interven- frequency and intensity might link to other
tion protocol. This can then be verified by forms of psychopathology (e.g., depression,
randomly selecting a percentage of sessions anxiety) and well-being. The newly developed
(e.g., 20%), with an independent researcher Compassion Engagement and Action Scale
observing or watching recorded sessions, offers a new approach in assessing compassion
with reliability statistics being reported (see motivation, which will be helpful for compas-
Moncher & Prinz, 1991, for examples on sion researchers (Gilbert et al., 2017).
how to assess treatment fidelity). 2. Improve specificity in measurement and com-
8. Collect follow-up data of at least 6 months, paring impacts of intervention when the target
preferably 12 months postintervention or of compassion is a family member, a familiar
longer. Only 50% of the studies in our meta- person, or a stranger, which could also be
analysis reported follow-up data. further assessed in terms of in- and out-group
9. Include clear descriptions of intervention length, variations (e.g., gender, race, ethnicity). This
and what is considered the minimum recom- level of specificity may further reveal the impact
mended dosage for interventions. It would of compassion-based interventions, and such
also be helpful to collect compliance data to knowledge would be of benefit when attempt-
determine how much meditation practice or ing to create compassionate schools, work-
other assigned homework exercises are being places, communities, and to foster a broader
conducted between intervention sessions. sense of global compassion.
10. Report attrition and include a CONSORT 3. There is a need to begin to assess the com-
flow diagram of participants in the study, and ponents of the intervention models to deter-
analyze data using intent-to-treat analyses. mine the mechanisms of change. We have
11. Minimize potential bias by preregistering trials initial evidence that these interventions are
on clinical databases (e.g., ClinicalTrials.gov working, but we need to start to understand
http://www.clinicaltrials.gov/) and include in more depth how these interventions are
conflict-of-interest statements. In this meta- working, by starting to examine the processes
analysis only 25% of studies included a COI of these interventions. One way to assist in
statement. this step would be to include more detail of
12. Begin to facilitate independent evaluations intervention components in supplementary
(e.g., intervention developer not included in material of published articles, for example,
the study) of compassion-based interventions. the transcripts of guided meditations.
This would help provide replication studies of 4. There is a need to investigate whether there
compassion interventions to determine their is an ideal intervention dosage, particularly in
reliability. regards to meditation length (e.g., length of
each individual meditation session or length of
future research the intervention program itself), to document
Based on this review, there are eight specific recom- dosage impacts. For example, does the amount
mendations we would like to make in order to of meditation required to reduce stress differ
improve understanding of the impacts of compassion- between clinical and nonclinical samples?
based interventions: 5. Conduct RCTs of compassion-based inter-
ventions with children or adolescents.
1. Current measures of compassion for self and 6. Assess for the acceptability of compassion-
others focus on trait measurement (e.g., SCS, based interventions. A key reason to assess for
Neff). What would be helpful is for compas- consumer acceptability of a program is that
sion measures to also assess the frequency individuals are more likely to access treat-
and intensity that people are experiencing ments that they view as acceptable (Borrego
compassion (for self, for others, from others) & Pemberton, 2007), while treatments that are
in their everyday lives. This is analogous to perceived as unacceptable may not be accessed
anxiety—all individuals experience some level regardless of their effectiveness (Eckert &
of anxiety, but the frequency and intensity of Hintze, 2000).
790 kirby et al.

7. Explore the potential impact of therapist or American Psychological Association (2008). Reporting stan-
teacher experience when delivering compas- dards for research in psychology: Why do we need them?
What might they be? American Psychologist, 63, 839–851.
sion interventions. For example, in compassion http://dx.doi.org/10.1037/0003-066X.63.9.839
interventions that are based on guided medita- *Arimitsu, K. (2016). The effects of a program to enhance
tions, is there a need for personal practice self-compassion in Japanese individuals: A randomized
similar to mindfulness-based interventions? controlled pilot study. The Journal of Positive Psychology,
(e.g., Crane et al., 2013). The issue of 11, 559–571. http://dx.doi.org/10.1080/17439760.2016.
1152593
self-practice/self-reflection is becoming increas- Bijmolt, T., & Pieters, R. (2001). Meta-analysis in marketing
ingly recognized as an important therapist when studies contain multiple measurements. Marketing
skill to enhance outcomes, and this needs to Letters, 12, 157–169.
be explored further (Thwaites et al., 2015). Borenstein, M., Hedges, L., Higgins, J., & Rothstein, H. (2005).
8. Meta-analyses of compassion-based interven- Comprehensive meta-analysis Version 2. Englewood, NJ:
Biostat.
tions would benefit from conducting moderator Borenstein, M., Hedges, L., Higgins, J. P. T., & Rothstein,
analyses on a range of variables, including, H. R. (2009). Introduction to Meta-Analysis. Hoboken,
but not limited to, intervention type (e.g., MSC, NJ: Wiley.
CEB), intervention length, intervention delivery Borrego, J., & Pemberton, J. R. (2007). Increasing acceptance
modality (e.g., group, individual), gender, age, of behavioural child management techniques: What do
parents say? Child & Family Behaviour Therapy, 29,
developer involvement, severity of initial pre- 27–43. http://dx.doi.org/10.1300/J019v29n02_03.
sentation of population sample, therapist/ Card, N. A. (2012). Applied meta-analysis for social science
teacher background (e.g., psychologist, teacher), research. New York, NY: The Guilford Press.
and unpublished papers. Future meta-analyses Carlson, K. D., & Schmidt, F. L. (1999). Impact of experimental
should examine the differences in effects of design on effect size: Findings from the research literature on
training. Journal of Applied Psychology, 84, 851–862.
RCTs comparing compassion interventions to *Carson, J. W., Keefe, F. J., Lynch, T. R., Carson, K. M., Goli, V.,
active control conditions versus waitlist condi- Fras, A. M., & Thorp, S. R. (2005). Loving-kindness meditation
tions (e.g., see Khoury et al., 2013). for chronic low back pain: Results from a pilot trial. Journal
of Holistic Nursing, 23, 287–304. http://dx.doi.org/10.1177/
conclusion 0898010105277651
This is the first meta-analysis documenting the im- Center for Compassion and Altruism Research and Education
(CCARE, 2015). Retrieved from http://ccare.stanford.edu/
pacts of compassion-based interventions. The results Cheung, M. W. L. (2013). Multivariate meta-analysis as
indicate that compassion-based interventions hold structural equation models. Structural Equation Modelling,
promise as a form of intervention to help cultivate 20, 429–454.
both compassion and self-compassion, reduce suffer- Cohen, J. (1992). A power primer. Psychological Bulletin, 112,
ing (specifically depression, anxiety, and psycholog- 155–159. http://dx.doi.org/10.1037/0033-2909.112.1.155
Crane, R. S., Eames, C., Kuyken, W., Hastings, R. P., Williams,
ical distress), as well as increase well-being. However, J. M. G., Bartley, T. . . . Surway, C. (2013). Development
the evidence base underpinning compassion-based and validation of the mindfulness-based interventions –
interventions relies predominantly on small sample teaching assessment criteria (MBI:TAC). Assessment, 20,
sizes. We have provided a series of recommendations 681–688. http://dx.doi.org/10.1177/1073191113490790
to help evaluators when designing RCTs aimed to Craske, M. G., Niles, A. N., Burkland, L. J., Wolitsky-Taylor,
K. B., Vilardaga, J. C. P. . . . Lieberman, M. D. (2014).
examine compassion-based interventions. RCT of CBT and ACT for social phobia: outcomes and
moderators. Journal of Consulting and Clinical Psychology,
Conflict of Interest Statement 82, 1034–1048. http://dx.doi.org/10.1037/a0037212
Crocker, J., & Canevello, A. (2012). Consequences of
The authors declare that there are no conflicts of interest. self-image and compassionate goals. In P. G. Devine, & A.
Plant (Eds.), Advances in experimental social psychology
(pp. 229–277). New York: Elsevier.
Appendix A. Supplementary Data Cuijpers, P., Cristea, I. A., Karyotaki, E., Reijnders, M., &
Supplementary data to this article can be found Huibers, M. J. H. (2016). How effective are cognitive behavior
online at http://dx.doi.org/10.1016/j.beth.2017.06. therapies for major depression and anxiety disorders A
003. meta-analytic update of the evidence. World Psychiatry, 15,
245–258. http://dx.doi.org/10.1002/wps.20346
*Desbordes, G., Negi, L. T., Pace, T. W. W., Wallace, B. A.,
References* Raison, C. L., & Schwartz, E. L. (2012). Effects of
mindful-attention and compassion mediation training on
*Albertson, E. R., Neff, K. D., & Dill-Shackleford, K. E. (2014). amygdala response to emotional stimuli in an ordinary,
Self-compassion and body dissatisfaction in women: A non-meditative state. Frontiers in Human Neuroscience, 6.
randomized controlled trial of a brief meditation intervention. http://dx.doi.org/10.3389/fnhum.2012.00292
Mindfulness. http://dx.doi.org/10.1007/s12671-014-0277-3 Duval, S., & Tweedie, R. (2000). Trim and fill: A simple funnel
plot based method of testing and adjusting for publication
⁎
Asterisk indicate studies included in the meta-analysis. bias in meta-analysis. Biometrics, 56, 455–463.
 compassion meta-analysis 791

Eckert, T. L., & Hintze, J. M. (2000). Behavioural conceptions and Psychology, 59, 12–19. http://dx.doi.org/10.1037/0022-
applications of acceptability: Issues related to service delivery 006X.59.1.12
and research methodology. School Psychology Quarterly, 15, *Jazaieri, H., Jinpa, T., McGonigal, K., Rosenberg, E. L.,
123–148. http://dx.doi.org/10.1037/h0088853. Finkelstein, J., Simon-Thomas, E., . . . Goldin, P. R. (2013).
Ekman, P. (2016). What scientists who study emotion agree Enhancing compassion: A randomized controlled trial of
about. Perspectives on Psychological Science, 11, 31–34. a compassion cultivation training program. Journal of
http://dx.doi.org/10.1177/1745691615596992 Happiness Studies, 14, 1113–1126. http://dx.doi.org/10.
*Feliu-Soler, A., Pascual, J. C., Elices, M., Martin-Blanco, A., 1007/s10902-012-9373-z
Carmona, C., Cebolla, A., Simon, V., & Soler, J. (2016). *Jazaieri, H., McGonigal, K., Jinpa, T., Doty, J. R., Gross, J. J.,
Fostering self-compassion and loving-kindness in patients & Goldin, P. R. (2014). A randomized controlled trial of
with borderline personality disorder: A ranomized pilot compassion cultivation training: Effects on mindfulness,
study. Clinical Psychology & Psychotherapy, 24, 278–286. affect, and emotion regulation. Motivation and Emotion,
http://dx.doi.org/10.1002/cpp.2000 38, 23–35. http://dx.doi.org/10.1007/s11031-013-9368-z
Fredrickson, B. L., Grewen, K. M., Coffey, K. A., Algoe, S. B., Jinpa, T. (2015). A Fearless Heart: How the Courage to be
Firestine, A. M., Arevalo, J. M. G., Ma, J., & Cole, S. W. Compassionate can Transform our Lives. New, York,
(2013). A functional genomic perspective on human New York: Hudson Street Press.
well-being. Proceedings of the National Academy of *Kelly, A. C., Zuroff, D. C., & Shapira, L. B. (2009). Soothing
Sciences, 110, 13684–13689. http://dx.doi.org/10.1073/ oneself and resisting self-attacks: The treatment of two
pnas.1305419110 intrapersonal deficits in depression vulnerability. Cognitive
*Friis, J., Johnson, M., Cutfield, R., & Consedine, N. (2016). Therapy and Research, 33, 301–313. http://dx.doi.org/10.
Kindness matters: A randomized controlled trial of a 1007/s10608-008-9202-1
mindful self-compassion intervention improves depression, Kemeny, M. E., Foltz, C., Cavanagh, J. F., Cullen, M.,
distress. Diabetes Care. http://dx.doi.org/10.2337/dc16- Giese-Davis, J., Jennings, P., . . . Ekman, P. (2012).
0416/-/DC1 Contemplative/emotion training reduces negative emotional
Gilbert, P. (1995). Biopsychosocial approaches and evolution- behavior and promotes prosocial responses. Emotion, 12,
ary theory as aids to integration in clinical psychology and 338–350. http://dx.doi.org/10.1037/a0026118
psychotherapy. Clinical Psychology and Psychotherapy, 2, Khoury, B., Lecomte, T., Fortin, G., Masse, M., Therien, P.,
135–156. http://dx.doi.org/10.1002/cpp.5640020302 Bouchard, V., . . . Hoffman, S. (2013). Mindfulness-
Gilbert, P. (2014). The origins and nature of compassion based therapy: A comprehensive meta-analysis. Clinical
focused therapy. British Journal of Clinical Psychology, 53, Psychology Review, 763–771. http://dx.doi.org/10.1016/
6–41. http://dx.doi.org/10.1111/bjc.12043 j.cpr.2013.05.005
Gilbert, P., Catarino, F., Duarte, C., Matos, M., Kolts, R., Kirby, J. N. (2016). Compassion interventions: the programmes,
Stubbs, J., Ceresatto, L., Duarte, J., Pinto-Gouveia, J., & the evidence, and implications for research and practice.
Basran, J. (2017). The development of compassionate Psychology and Psychotherapy: Theory, Research and
engagement and action scales for self and others. Journal Practice. http://dx.doi.org/10.1111/papt.12104
of Compassionate Health Care, 4, 4. http://dx.doi.org/10. Kirby, J. N., Doty, J., Petrocchi, N., & Gilbert, P. (2017).
1186/s40639-017-0033-3 The current and future role of heart rate variability for
Gleser, L. J., & Olkin, I. (2007). Stochastically dependent effect assessing and training compassion. Frontiers in Public Health,
sizes (technical report 2007 - 2). http://statistics.stanford. 5, 40. http://dx.doi.org/10.3389/fpubh.2017.00040.
edu/~ckirby/techreports/GEN/2007/2007-2.pdf (Retrieved Kirby, J. N., & Gilbert, P. (2017). The emergence of the
from Stanford University website) compassion focused therapies. In P. Gilbert (Ed.), Compassion:
Goetz, J. L., Keltner, D., & Simon-Thomas, E. (2010). concepts, research and applications (pp. 258–285). London:
Compassion: An evolutionary analysis and empirical review. Routledge.
Psychological Bulletin, 136, 351–374. http://dx.doi.org/10. Kirby, J. N., Tellegen, C. L., & Steindl, S. R. (2015).
1037/a0018807 Cultivating compassion: A systematic review and meta-
Hangartner, D. (2013). Human suffering and the four analysis of compassion-based interventions. PROSPERO:
immeasurables – A Buddhist perspective on compassion. In International Prospective Register of Systematic Reviews,
T. Singer & M. Bolz (Eds.), Compassion: Bridging Practice 2015, CRD42015024576. Retrieved from http://www.crd.
and Science (pp. 152–164). Munich: Max Planck Society. york.ac.uk/PROSPERO/display_record.asp?ID=
Hedges, L., & Olkin, I. (1985). Statistical methods for meta- CRD42015024576
analysis. San Diego, CA: Academic Press. Klimecki, O. M., Leiberg, S., Ricard, M., & Singer, T. (2014).
*Held, P., & Owens, G. P. (2015). Effects of self-compassion Differential pattern of functional brain plasticity after
workbook training on trauma-related guilt in a sample compassion and empathy training. Social Cognitive &
of homeless veterans: A pilot study. Journal of Clinical Affective Neuroscience, 7, 9, 873–879. http://dx.doi.org/
Psychology. http://dx.doi.org/10.1002/jclp.22170 10.1093/scan/nst060
Higgins, J. P. T., Altman, D. G., Gotzsche, P. C., Juni, P., Leaviss, J., & Uttley, L. (2015). Psychotherapeutic benefits of
Moher, D., Oxman, A. D., . . . Sterne, J. A. C. (2011). compassion-focused therapy: An early systematic review.
The Cochrane Collaboration’s tool for assessing risk of bias Psychological Medicine, 45, 927–945. http://dx.doi.org/
in randomised trials. British Medical Journal, 343, d5928. 10.1017/S0033291714002141
http://dx.doi.org/10.1136/bmj.d5928 *Lee, W. K., & Bang, H. J. (2010). The effects of
Huedo-Medina, T. B., Sanchez-Meca, J., Marin-Martinez, F., & mindfulness-based group intervention on the mental health
Botella, J. (2006). Assessing heterogeneity in meta-analysis: of middle-aged Korean women in community. Stress and
Q statistic or I2 index? Psychological Methods, 11, 193–206. Health: Journal of the International Society for the
http://dx.doi.org/10.1037/1082-989X.11.2.193 Investigation of Stress, 26, 341–348. http://dx.doi.org/10.
Jacobson, N. S., & Truax, P. (1991). Clinical significance: 1002/smi.1303
A statistical approach to defining meaningful change in Leiberg, S., Klimecki, O., & Singer, T. (2011). Short-term
psychotherapy research. Journal of Consulting and Clinical compassion training increases prosocial behavior in a
792 kirby et al.

newly developed prosocial game. PLoS One, 6(3), e17798. Pace, T. W., Negi, L. T., Adame, D. D., Cole, S. P., Sivilli, T. I.,
http://dx.doi.org/10.1371/journal.pone.0017798 Brown, T. D., . . . Raison, C. L. (2009). Effect of compassion
Liberati, A., Altman, D. G., Tetzlaff, J., Mulrow, C., Götzsche, meditation on neuroendocrine, innate immune and behavioral
P. C., Ioannidis, J. P. A., . . . Moher, D. (2009). The PRISMA responses to psychosocial stress. Psychoneuroendocrinology,
statement for reporting systematic reviews and meta- 34, 87–98. http://dx.doi.org/10.1016/j.psyneuen.2008.08.011
analyses of studies that evaluate health care interventions: *Pons, G.B. (2014). Cultivating healthy minds and open hearts:
Explanation and elaboration. PLoS Medicine, 6, A mixed-method controlled study on the psychological and
e1000100. http://dx.doi.org/10.1371/journal.pmed. relational effects of compassion cultivation training in Chile.
1000100. Unpublished doctoral dissertation.
*Lo, H. H., Ng, S. M., Chan, C. L., Lam, K. F., & Lau, B. H. Sanders, M. R., & Kirby, J. N. (2014). Surviving or thriving:
(2013). The Chinese medicine construct “stagnation” in Quality assurance mechanisms to promote innovation in
mind-body connection mediates the effects of mindfulness the development of evidence- based parenting interventions.
training on depression and anxiety. Complementary Prevention Science. Advanced publication. http://dx.doi.org/
Therapies in Medicine, 21, 348–357. http://dx.doi.org/10. 10.1007/s11121-014-0475-1
1016/j.ctim.2013.05.008 Seppala, E., Rossomando, T., & Doty, J. R. (2013). Social
MacBeth, A., & Gumley, A. (2012). Exploring compassion: connection and compassion: Important predictors of health
A meta-analysis of the association between self-compassion and well-being. Social Research, 80, 411–430.
and psychopathology. Clinical Psychology Review, 32, *Shahar, B., Szsepsenwol, O., Zilcha-Mano, S., Haim, N.,
545–552. http://dx.doi.org/10.1016/j.cpr.2012.06.003 Zamir, O., Levi-Yeshuvi, S., & Levit-Binnun, N. (2015).
*Mantzios, M., & Wilson, J. C. (2014). Making concrete A wait-list randomized controlled trial of loving-kindness
construals mindful: A novel approach for developing meditation programme for self-criticism. Clinical Psychology
mindfulness and self-compassion to assist weight loss. & Psychotherapy, 22, 346–356. http://dx.doi.org/10.1002/
Psychology & Health, 29, 422–441. http://dx.doi.org/10. cpp.1893
1080/08870446.2013.863883 Singer, T., & Bolz, M. (2013). Compassion: Bridging practice
Marin-Martinez, F., & Sanchez-Meca, J. (1999). Averaging and science. Munich: Max Planck Society.
dependent effect sizes in metaanalysis: A cautionary note *Smeets, E., Neff, K., Alberts, H., & Peters, M. (2014). Meeting
about procedures. The Spanish Journal of Psychology, 2, suffering with kindness: Effects of a brief self-compassion
32–38. intervention for female college students. Journal of Clinical
Mascaro, J. S., Rilling, J. K., Nego, L. T., & Raison, C. L. Psychology, 70, 794–807. http://dx.doi.org/10.1002/jclp.
(2013). Compassion meditation enhances empathic accura- 22076
cy and related neural activity. Social Cognitive and Affective Strauss, C., Taylor, B. L., Gu, J., Kuyken, W., Baer, R., Jones,
Neuroscience, 8, 48–55. http://dx.doi.org/10.1093/scan/ F., & Cavanagh, K. (2016). What is compassion and how
nss095 can we measure it? A review of definitions and measures.
Moher, D., Hopewell, S., Schulz, K. F., Montori, V., Gotzsche, Clinical Psychology Review, 47, 15–27. http://dx.doi.org/
P. C., Devereaux, P. J., . . . Altman, D. G. (2010). 10.1016/j.cpr.2016.05.004
CONSORT 2010 explanation and elaboration: Updated Thwaites, R., Cairns, L., Bennett-Levy, J., Johnson, L., Lowrie,
guidelines for reporting parallel group randomized trials. R., Robinson, A., Turner, M., Haarhoff, B., & Perry, H.
BMJ, 340. http://dx.doi.org/10.1136/bmj.c869 (2015). Developing metacompetence in low intensity
Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). cognitive- behavioural therapy (CBT) interventions: Evaluating
Preferred reporting items for systematic reviews and a self-practice/self-reflection programme for experienced
meta-analyses: The PRISMA statement. PLoS Medicine, 6, low intensity CBT practitioners. Australian Psychologist, 50,
e1000097. 311–321. http://dx.doi.org/10.1111/ap.12151
Moncher, F., & Prinz, R. J. (1991). Treatment fidelity in *Toole, A. M., & Craighead, L. W. (2016). Brief self-compassion
outcome studies. Clinical Psychology Review, 11, 247–266. meditation training for body image distress in young adult
http://dx.doi.org/10.1016/0272-7358(91)90103-2 women. Body Image, 19, 104–112. http://dx.doi.org/10.1016/
Morris, S. B. (2008). Estimating effect sizes from pretest-posttest- j.bodyim.2016.09.001
control group designs. Organizational Research Methods, 11, *Wallmark, E., Safarzadeh, K., Daukantaitė, D., & Maddux,
364–386. http://dx.doi.org/10.1177/1094428106291059 R. E. (2013). Promoting altruism through meditation: An
*Mosewich, A. D., Crocker, P. R. E., Kowalski, K. C., & 8-week randomized controlled pilot study. Mindfulness, 4,
DeLongis, A. (2013). Applying self-compassion in sport: 223–234. http://dx.doi.org/10.1007/s12671-012-0115-4
An intervention with women athletes. Journal of Sport & *Weibel, D. T. (2008). A loving-kindness intervention: Boosting
Exercise Psychology, 35, 514–524. compassion for self and others (Dissertation). ProQuest
Neff, K. (2003). The development and validation of a scale to Information & Learning, US.
measure self-compassion. Self and Identity, 2, 223–250. Yarnell, L. M., & Neff, K. D. (2013). Self-compassion, interper-
http://dx.doi.org/10.1080/15298860390209035. sonal conflict resolutions, and well-being. Self and Identity, 12,
*Neff, K. D., & Germer, C. K. (2013). A pilot study and 146–159. http://dx.doi.org/10.1080/15298868.2011.649545
randomized controlled trial of the mindful self-compassion
program. Journal of Clinical Psychology, 69, 28–44.
http://dx.doi.org/10.1002/jclp.21923 R E C E I V E D : June 22, 2016
Orwin, R. G. (1983). A fail-safe N for effect size in meta- A C C E P T E D : June 15, 2017
analysis. Journal of Educational Statistics, 8, 157–159. A V A I L A B L E O N L I N E : 21 June 2017