Behaviour Research and Therapy 146 (2021) 103965

Contents lists available at ScienceDirect

Behaviour Research and Therapy

journal homepage: www.elsevier.com/locate/brat

The effects of inducing self-compassion on affective and physiological

recovery from a psychosocial stressor in depression
Alison Tracy *, Ellen Jopling, Joelle LeMoult
The University of British Columbia, 2136 West Mall, Vancouver, BC, V6T1Z4, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: The current study was designed to extend previous research by testing whether self-compassion acts as a pro­
Depression tective factor that facilitates faster affective and physiological recovery from stress in people with elevated
Self-compassion depressive symptoms. Specifically, we examined the effect of experimentally induced self-compassion on positive
Affect
affect, negative affect, and respiratory sinus arrhythmia (RSA) recovery from stress. Participants (N = 59)
Respiratory sinus arrhythmia
Stress
experiencing elevated depressive symptoms completed the Trier Social Stress Test (TSST), a standardized psy­
chosocial stressor, and then were randomly assigned to either a self-compassion induction or a no-strategy
control induction before resting quietly during the 30-min recovery period. During the induction period, par­
ticipants in the self-compassion condition exhibited a greater increase in positive affect and a trend towards a
greater decrease in negative affect than did participants in the no-strategy control condition. However, the
psychological benefits of self-compassion did not continue during the post-induction recovery period. Moreover,
changes in RSA levels did not differ between participants in the self-compassion and no-strategy control con­
dition. These results suggest that, among individuals with elevated depressive symptoms, brief self-compassion
inductions have short-term beneficial psychological, but not physiological, effects. As such, our findings delin­
eate the benefits and boundaries of single-session self-compassion inductions in depression, and in doing so,
inform future experimental and applied research.

1. Introduction suffers and identifying with universal suffering; and (3) mindfulness –
being aware of painful thoughts and feelings without over-thinking
An extensive body of research documents the role of stress in the them. The construct of self-compassion has been proposed as an adap­
onset, maintenance, and recurrence of depression (Hammen, 2005). Yet, tive means of relating to oneself (Neff, Kirkpatrick, & Rude, 2007).
despite the frequency with which stressful life events typically occur Further, it has been suggested that self-compassion may be a resiliency
(Ozer, Best, Lipsey, & Weiss, 2003), a relatively small subset of the factor that facilitates adaptive stress recovery and, thus, protects against
population experiences clinically significant depressive symptoms the development and maintenance of depression during times of stress
(Kessler et al., 2003). Researchers have shown that the way individuals (Ehret, Joormann, & Berking, 2015).
regulate their emotions following stress plays a central role in deter­ Consistent with this formulation, numerous studies have demon­
mining their emotional recovery, which is vital in protecting against the strated an inverse association between self-compassion and symptoms of
development and exacerbation of depression (Coifman & Bonanno, depression, with a recent meta-analysis finding a large mean effect size
2010; Troy, Wilhelm, Shallcross, & Mauss, 2010). It is, therefore, critical (MacBeth & Gumley, 2012). Further, multiple studies assessing partic­
to examine factors that could promote effective recovery from stress. ipants with depression have shown that they possess lower levels of
One factor that has been increasingly examined in connection to self-compassion than never-depressed controls (Ehret et al., 2015;
stress recovery and depression is self-compassion, a construct with a Krieger, Altenstein, Baettig, Doerig, & Holtforth, 2013). Moreover,
long tradition in Buddhist teachings. There are three core facets of self- although experimental studies examining the association between
compassion as defined by Neff (2003a): (1) self-kindness – being kind self-compassion and affect in the context of depression are sparse, the
and understanding to oneself instead of being judgmental and two studies that have been conducted both found that participants
self-critical; (2) common humanity – understanding that everyone assigned to a self-compassion induction reported significantly reduced

* Corresponding author. 2136 West Mall, Department of Psychology, University of British Columbia, Vancouver, BC, V6T1Z4, Canada.
E-mail address: alison.tracy@psych.ubc.ca (A. Tracy).

https://doi.org/10.1016/j.brat.2021.103965
Received 28 July 2020; Received in revised form 12 December 2020; Accepted 7 September 2021
Available online 11 September 2021
0005-7967/© 2021 Elsevier Ltd. All rights reserved.
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

depressed mood compared to those assigned to a no-strategy control 2. Methods

condition (Diedrich, Grant, Hofmann, Hiller, & Berking, 2014; Ehret,
Joormann, & Berking, 2018). 2.1. Participants
Given the documented benefits of self-compassion on depressed
mood (Diedrich et al., 2014; MacBeth & Gumley, 2012), the goal of the Participants were 59 adults with elevated symptoms of depression.
current study was to test the boundaries of the benefits of Participants were recruited from flyers posted in the community, online
self-compassion in depression by considering other outcome measures. advertisements posted on community forums, and the student research
In recent years, research on emotion has increasingly recognized the pool at the University of British Columbia. Individuals participated in
importance of assessing multiple outcome measures (Scherer, 2004), the study in exchange for a monetary honorarium or, if they were
and this has been recommended with regards to self-compassion spe­ recruited through the student research pool, for course credit. In­
cifically (Ehret et al., 2018). For example, theoretical models concep­ dividuals were eligible if they were between the ages of 18–65 inclusive,
tualize negative and positive affect as distinct and orthogonal constructs were fluent in English, and passed a two-step screening process that has
(Tellegen, Watson, & Clark, 1999) that are associated with separate been used in other studies recruiting depressed participants (e.g.,
domains in the Research Domain Criteria (RDoC) framework (Insel Hakstian & McLean, 1989). Specifically, participants who endorsed at
et al., 2010), and researchers have documented that a lack of positive least 5 current symptoms of major depressive disorder from the Diag­
affect is an important and independent predictor of depression (e.g., nostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5)
Gentzler & Root, 2019; Kuhlman et al., 2019). Further, researchers are in an online screener (American Psychiatric Association, 2013) were
increasingly incorporating physiological markers of stress to gain a more invited to the laboratory to complete the Beck Depression Inventory –
comprehensive picture of the stress response. Cardiac vagal control is a Second Edition (BDI-II; Beck, Steer, & Brown, 1996), and those who
central physiological marker of stress that has been implicated in risk for reported mild to severe symptoms of depression on the BDI (i.e., BDI-II
depression (Rottenberg, 2007). Cardiac vagal control is often quantified scores ≥ 14 as recommended by Beck et al., 1996) were eligible to
by respiratory sinus arrhythmia (RSA), a measure of variability in heart participate in the study. The mean age of participants was 20.20 years
rate that occurs over the respiration cycle (Porges, 2007). RSA is an (SD = 2.56). The majority of participants self-identified as Asian (53%),
important construct to examine as RSA withdrawal is posited to facili­ while 25% identified as White, 3% as Hispanic/Latino, and 19% as other
tate an individuals’ ability to cope with stress by mediating metabolic ethnicities.
output to increase heart rate (Porges, 2007). With this in mind, RSA is
typically highest during periods of rest, decreases rapidly in times of 2.2. Procedure
stress, and then increases to facilitate autonomic recovery (Kreibig,
2010). Depression has been associated with attenuated RSA recovery The present study was approved by the Behavioural Research Ethics
following stressor offset (Rottenberg, Wilhelm, Gross, & Gotlib, 2003), Board at the University of British Columbia. After providing informed
which is important because prolonged RSA recovery from stress has consent, participants completed the BDI-II to determine eligibility.
been linked to adverse cardiac outcomes (Crowell, Skidmore, Rau, & Eligible participants then watched a 15-min nature video to assess
Williams, 2013). baseline levels of affect and RSA. Next, participants completed a stan­
Despite the potential benefits of identifying constructs, like self- dard psychosocial stressor, the TSST (Kirschbaum et al., 1993). Imme­
compassion, that facilitate affective and RSA recovery from stress, few diately following the stressor, participants were randomly assigned to
studies have done so. In fact, the only two experimental studies exam­ complete either a self-compassion induction or no-strategy control in­
ining the association between self-compassion and affect in the context duction. Random assignment was conducted using an online random
of depression measured negative, but not positive, affect, and neither number generator to independently assign each participant the number
included physiological markers of stress (Diedrich et al., 2014; Ehret 1 or 2, which categorized them into the self-compassion or no-strategy
et al., 2018). control condition. Experimenters were blind to participants’ condition.
The current study was designed to extend previous research by To examine affective and RSA levels following induction offset, partic­
testing whether self-compassion acts as a protective factor that facili­ ipants then completed an unstructured recovery period, during which
tates faster affective and physiological recovery from stress in people they sat quietly for 30 min without any distractors. Affect and RSA were
with elevated depressive symptoms. Specifically, we examined the effect collected throughout (as described in the ‘Measures’ section).
of experimentally induced self-compassion on positive affect, negative
affect, and RSA recovery from stress. Consistent with past research (Lee, 2.3. Psychosocial stressor
Mathews, Shergill, & Yiend, 2016; Pictet, Jermann, & Ceschi, 2016), we
recruited a sample of individuals who endorsed elevated depressive To induce a moderate amount of stress, participants completed the
symptoms as measured by the Beck Depression Inventory- Second Edi­ TSST (Kirschbaum et al., 1993), a standardized and well-validated
tion (BDI-II). Participants completed the Trier Social Stress Test (TSST), psychosocial stressor. Immediately following the baseline nature
a standardized psychosocial stressor (Kirschbaum, Pirke, & Hell­ video, participants were told they would give a 5-min speech to com­
hammer, 1993). They were then randomly assigned to either a mittee members, who would rate their speech quality. Participants had
self-compassion induction or a no-strategy control induction. Although 3 min to prepare their speech, after which time three confederates (two
past research on self-compassion has used a within-subject design where females, one male) entered the room. Participants then gave their
each participant completes each induction, carry-over effects from one speech, during which the confederates maintained a neutral facial
induction to the next have been documented (Diedrich, Hofmann, expression and provided no feedback. The male confederate asked
Cuijpers, & Berking, 2016). Thus, we used a between-subject design. scripted questions if time remained. After the speech, participants
Participants’ levels of affect and RSA were measured throughout the completed an unexpected 5-min mental math task, in which they serially
study. We expected that the self-compassion induction would be subtracted the number 13 from 1022. Participants were told that their
significantly more effective than the no-strategy control condition at performance was being video recorded in line with standard TSST
promoting recovery from stress, as indicated by measures of negative protocol.
affect, positive affect, and RSA.
2.4. Stress-response (SR) induction

After completing the stressor, participants were randomly assigned

(as described above) to receive one of two stress-response (SR)

2
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

inductions: self-compassion or a no-strategy control condition. The SR 2.6. Planned analyses

inductions were provided via pre-recorded audio files to ensure con­
sistency across participants, and participants listened to them through Affect and physiological data were analyzed using multilevel
headphones connected to a portable MP3 player. Immediately after modeling, which allowed us to model at Level 1 repeated measurements
participants began listening to the audio file, the experimenter left the of positive affect, negative affect, and RSA within persons as a function
room; thus, participants listened to the induction while sitting alone in a of time (Raudenbush & Bryk, 2002). Multilevel modeling is ideal for
private room with no distractions. Both SR conditions began the same modeling the nested structure of the data as it does not assume inde­
way, with participants being instructed to think about the kinds of pendence of data points and handles varying time intervals between
thoughts and feelings they were experiencing as a result of the TSST measurements (Hruschka, Kohrt, & Worthman, 2005). For each
speech and math task that they had just completed. In the self- outcome, we evaluated linear, quadratic, and piecewise models; we then
compassion induction, participants then received instructions intended selected the model that best fit the data based on deviance statistics,
to help them achieve a more self-compassionate perspective of their visual inspection of the data, and the smallest value of Akaikie’s Infor­
performance during the TSST. The self-compassion instructions were mation Criteria (AIC). Next, we tested a series of potential covariates
adapted from a guided self-compassion meditation available at that, based on previous research, might influence responses to stress:
self-compassion.org (Neff, 2016) and were consistent with past research age, sex, body mass index (BMI), current use of psychotropic medica­
on self-compassion inductions (e.g., Diedrich et al., 2014). For example, tion, and current use of oral contraceptives (Anderl, Li, & Chen, 2020;
participants were instructed to give themselves the same kindness, Jorm, 1987; Moncrieff, Cohen, & Porter, 2013; O’Regan, Kenny, Cronin,
support, and compassion that they would give to a friend who was Finucane, & Kearney, 2015; Speed, Jefsen, Børglum, Speed, &
feeling the same way. In contrast, in the no-strategy control condition, Østergaard, 2019; Tonhajzerova et al., 2008; Voss, Schroeder, Heit­
participants were instructed to sit in silence for the remainder of the mann, Peters, & Perz, 2015). We then tested whether between-person
induction (as is consistent with Diedrich et al., 2014 and Ehret et al., variability in Level 1 parameters was explained at Level 2 by charac­
2018). Both conditions lasted a total of 8 min. teristics that varied across individuals, namely participants’ assigned SR
condition, with relevant covariates included in the model. Analyses
2.5. Measures were run using hierarchical linear modeling software (HLM-7; Rau­
denbush, Bryk, & Congdon, 2004). Models were fit using full informa­
Positive and Negative Affect. Self-reported affect was assessed tion maximum likelihood for calculating deviance estimates and AIC
using an adapted version of the Positive and Negative Affect Schedule and using restricted maximum likelihood to estimate model parameters.
(PANAS; Watson, Clark, & Tellegen, 1988). Affect was assessed at 5 It has been recommended that for hierarchical models, Level 2 sample
time-points: following a 15-min nature video, after the psychosocial sizes should be greater than 50 to achieve adequate power (Maas & Hox,
stressor, after the stress-response induction, and twice during the re­ 2005). Further, a power analysis based on a medium effect size for
covery period (15- and 30-min post-SR induction). Consistent with past self-compassion (Cohen’s f = .15) indicated that a sample of 58 partic­
research (Waugh, Muhtadie, Thompson, Joormann, & Gotlib, 2012) ipants would be required to detect the hypothesized effect at α = .05 and
negative affect was calculated based on participants’ ratings of angry, power = 80% (Soper, 2020). Robust standard errors were used for all
anxious, and upset, α = .626; and positive affect was calculated based on analyses to reduce bias, following recommendations put forth by Rau­
participants’ ratings of inspired, happy, and interested, α = .707. denbush and Bryk (2002). All tests of significance were conducted using
Respiratory Sinus Arrhythmia (RSA). RSA was assessed by col­ two-tailed testing.
lecting Electrocardiograph (ECG) and cardiac impedance using Modelling Affect. For both positive and negative affect, the smallest
frequency-domain analysis. Physiological activity was recorded AIC value was associated with the piecewise model (AIC = 1154.77 and
continuously at a sampling rate of 1 kHz using a MindWare Mobile data 1205.78, respectively), which fit the data significantly better than both
acquisition device and Biolab Acquisition Software. To measure ECG, the linear and quadratic models, ps < .001. The piecewise model esti­
three standard electrodes were attached bilaterally to participants’ left mated affect at baseline (immediately following the 15-min nature
and right lower rib cage and right collarbone. To measure impedance, video), and during the stress reactivity (baseline to immediately after
one electrode was placed on the participant’s jugular notch and one was the stressor), SR induction (immediately after the stressor to immedi­
placed just below the sternum on the zyphoid process. Two electrodes ately after the SR induction), and recovery periods (immediately after
were placed on the back of the body. The cardiac impedance signal (Z0) the SR induction to 30 min post-induction). Thus, we specified the
is used to validate the RSA data by ensuring that the detected respiration following Level 1 models (one for positive affect and one for negative
rate falls within the high-frequency band (0.15–0.4 Hz). Data were affect):
analyzed using MindWare’s BioLab analysis software in 60-s increments. Affect = π0j (baseline)+ π1j (stress reactivity) + π2j (SR induction) +
The ECG signal was inspected for artifacts and missing R-peaks (based π3j (recovery) + eij
on improbable interbeat intervals) and were manually corrected. In this equation, π0j represents the level of positive/negative affect
Minute-by-minute estimates of RSA were determined for the baseline for participant j at baseline, π1j represents the slope of change in posi­
period (final 5 min of the nature video), the preparatory period (3 min), tive/negative affect during the stressor for participant j (positive values
the Trier stress period (10 min), the SR induction (8 min), and the re­ indicate an increase in affect during the stressor; higher values indicate a
covery period (first 5 min), as is consistent with past research (LeMoult, steeper slope), π2j represents the slope of change in positive/negative
Yoon, & Joormann, 2016). affect during the SR induction period for participant j (positive values
Questionnaires. Participants completed the BDI-II (Beck et al., indicate an increase in affect during the SR induction; higher values
1996), a 21-item measure used to assess depression severity. The BDI-II indicate a greater increase), π3j represents the slope of change in posi­
was normed on a clinical sample of depressed participants and has tive/negative affect across the recovery period for participant j (negative
shown good reliability and validity (Beck et al., 1996). The BDI-II values indicate a decrease in affect during the recovery period; lower
showed good internal reliability in this sample, α = .792. Participants values indicate a greater decrease in affect), and eij represents the
also completed the Self-Compassion Scale – Short Form (SCS-SF; Raes, within-person random effect for participant j.
Pommier, Neff, & Van Gucht, 2011) to assess baseline levels of We next tested as potential covariates the series of variables
self-compassion. The SCS-SF includes 12 of the original 26 items (Neff, described above: age, sex, BMI, current use of psychotropic medication,
2003b) but has a near-perfect correlation with the original measure and current use of oral contraceptives. For positive affect, we found that
(Raes et al., 2011). The SCS-SF scale showed good internal reliability sex was associated with change in positive affect in response to the
within the present sample, α = .79. stressor, p = .019, and both sex and current use of psychotropic

3
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

medication were associated with change in positive affect across the Table 1
induction period, p = .024 and p = .007, respectively. Thus, we included Participant characteristics.
these covariates in the corresponding Level 2 model. We also included Variable Control Condition Self-Compassion
condition in all Level 2 equations to examine whether assigned SR (n = 28) Condition (n = 31)
condition was associated with individual differences in affect: Age, M (SD) 20.36 (2.66) 20.06 (2.50)
Baseline Affect: π0j = B00 + B01 (condition) + r0. Proportion Female 86% 87%
Stress Reactivity: π1j = B10 + B11 (condition) + B12 (sex) + r1. BDI score, M (SD) 27.18 (7.48) 25.16 (8.19)
SR Induction: π2j = B20 + B21 (condition) + B22 (sex) + B23 SCS score, M (SD) 2.35 (0.64) 2.45 (0.62)
BMI, M (SD) 22.71 (3.59) 24.34 (6.15)
(psychotropic medication) + r2. Proportion taking 48% 26%
Recovery: π3j = B30 + B31 (condition) + r3. psychotropic medication
Of the aforementioned covariates, sex was associated with levels of Proportion taking oral 14% 23%
baseline negative affect, p = .001. Thus, for negative affect, we included contraceptives
Race
sex in the corresponding Level 2 model alongside condition:
Aboriginal and White 0% 7%
Baseline Affect: π0j = B00 + B01 (condition) + B02 (sex) + r0. Asian 64% 42%
Stress Reactivity: π1j = B10 + B11 (condition) + r1. Asian and White 7% 0%
SR Induction: π2j = B20 + B21 (condition) + r2. Asian and Other 0% 3%
Recovery: π3j = B30 + B31 (condition) + r3. Black and White 4% 0%
White 21% 29%
Modelling RSA. For RSA, the smallest AIC was associated with the White and Hispanic or 0% 3%
piecewise model (AIC = 798.72), which estimated RSA stress reactivity Latino
(from stress onset to stress offset), SR induction change (induction onset Other 3% 16%
to induction offset), and RSA recovery (initial 5 min of the recovery Household Income
than $30,000 19% 15%
period), which is in line with previous research (Arch et al., 2014;
Between $30,000 & $50,000 7% 8%
LeMoult et al., 2016), which fit the data significantly better than both Between $50,001 & $70,000 11% 12%
the linear and quadratic models, ps < .001. Therefore, we specified the Between $70,001 & $90,000 19% 15%
following Level 1 model: Between $90,001 & 7% 15%
RSA = π0j (baseline) + π1j (RSA reactivity) + π2j (induction change) $110,000
Between $110,001 & 15% 23%
+ π3j (RSA recovery) + eij
$130,000
In this equation, π0j represents the level of RSA for participant j at Greater than $130,001 22% 12%
baseline, π1j represents the slope of RSA reactivity for participant j, π2j
Note. BDI = Beck Depression Inventory; SCS = Self-Compassion Scale.
represents the slope of RSA for participant j across the induction period,
π3j represents the slope of RSA recovery for participant j, and e1j rep­
resents the within-person random effect for participant j. We then tested zero at baseline, B = 3.77, t (58) = 11.83, p < .001, and significantly
a series of variables that have been found to affect RSA in past studies as decreased in response to stress onset, B = − 0.11, t (58) = − 6.97, p <
potential covariates: age, sex, body mass index (BMI), current use of .001. Across all participants, level of positive affect remained stable
psychotropic medication, and current use of oral contraceptives across the induction period, B = 0.02, t (58) = 0.75, p = .458, and
(O’Regan et al., 2015; Tonhajzerova et al., 2008; Voss et al., 2015). Of significantly decreased across the recovery period, B = − 0.02, t (58) =
these variables, current use of psychotropic medication predicted RSA − 2.21, p = .031.
levels at baseline, p = .013; thus, it was included as a covariate in the We then added condition at Level 2 in order to examine whether
corresponding Level 2 model. We also included condition in all Level 2 individual differences in the change in positive affect across the psy­
equations to examine whether assigned SR condition was associated chosocial stressor were explained by assigned condition. Importantly,
with individual differences in RSA: condition did not predict baseline levels of positive affect, B = − 0.59, t
Baseline RSA: π0j = B00 + B01 (condition) + B02 (psychotropic (57) = − 0.94, p = .353, or changes in positive affect during the stressor,
medication) + r0 B = 0.04, t (57) = 1.16, p = .252, suggesting that random assignment
RSA Reactivity: π1j = B10 + B11 (condition) + r1 was effective. However, as expected, change in positive affect differed
Induction Change: π2j = B20 + B21 (condition) + r2 significantly by condition during the SR induction, B = 0.10, t (57) =
RSA Recovery: π3j = B30 + B31 (condition) + r3 2.48, p = .016, and recovery period, B = − 0.04, t (57) = − 2.31, p = .025.
For participants in the self-compassion condition, positive affect
3. Results increased during the self-compassion induction, B = 0.16, t (28) = 2.20,
p = .037, and then decreased following offset of the induction, B =
3.1. Demographic and clinical characteristics of participants by condition − 0.04, t (30) = − 2.73, p = .010. In contrast, for participants in the no-
strategy condition, positive affect did not change during the SR induc­
Descriptive and clinical characteristics of participants randomly tion, B = − 0.04, t (27) = − 1.71, p = .098, or recovery period, B =
assigned to the self-compassion and control conditions are presented by − 0.0001, t (27) = − 0.01, p = .994.
condition in Table 1. There were no significant differences between the We then conducted exploratory analyses to examine whether levels
two conditions on age, t (57) = 0.44, p = .665, proportion woman, χ 2 (1, of baseline self-compassion moderated the effect of condition on levels
N = 59) = 0.03, p = .877, levels of depression, t (57) = 0.98, p = .330, of positive affect across the psychosocial stressor by adding baseline self-
race, χ 2 (7, N = 59) = 10.95, p = .141, or household income, χ 2 (6, N = compassion scores and the interaction between baseline self-compassion
53) = 2.27, p = .893. and condition to all Level 2 equations. This analysis yielded a significant
interaction between self-compassion and condition during the SR in­
3.2. Effect of condition on positive and negative affect duction, B = 0.12, t(51) = 2.05, p = .045. Follow-up tests indicated that,
among those in the self-compassion condition, higher levels of baseline
Positive affect. Positive affect across the psychosocial stressor is self-compassion were associated with a greater increase in positive affect
presented by condition in Fig. 1. To examine the basic pattern of positive during the SR induction, B = 0.12, t (26) = 2.66, p = .013, suggesting
affect across the psychosocial stress task, we first ran a baseline model that the self-compassion induction was more effective for those with
without any predictors at Level 2. This model indicated that partici­ higher levels of baseline self-compassion. In contrast, within the no-
pants’ average level of positive affect was significantly different from strategy control condition, levels of baseline self-compassion were not

4
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

Fig. 1. Positive affect ratings for participants in the control and self-compassion conditions. Error bars indicate ± 1 standard error.

associated with change in positive affect during the SR induction, B = 10.22, p < .001, significantly decreased across the induction period, B =
0.02, t (23) = 0.64, p = .531. − 0.15, t (58) = − 6.55, p < .001, and significantly decreased across the
We also conducted exploratory analyses on each positive affective recovery period, B = − 0.06, t (58) = − 5.51, p < .001.
state separately. Positive affect was comprised of participants’ ratings of Participants in the self-compassion condition and the no-strategy
inspired, happy, and interested. As expected, condition did not predict control condition did not differ in negative affect at baseline, B =
baseline levels or stress-reactivity changes in any positive affect state, ps 0.38, t (56) = 1.12, p = .267, or in negative affective reactivity during
> .272. However, during the SR induction, increases in inspiration and the stressor, B = − 0.01, t (57) = − 0.31, p = .758, providing further
happiness were greater in the self-compassion condition than the no- support that random assignment was effective. However, as expected,
strategy control condition, ps < .025. Subsequent decreases in inspira­ decreases in negative affect during the SR induction were greater in the
tion and happiness during the recovery period were also greater in the self-compassion condition compared to the no-strategy control condi­
self-compassion condition than the no-strategy control condition, ps < tion at a trend level, B = − 0.09, t (57) = − 1.96, p = .055, with decreases
.048. In contrast, levels of interest during the psychosocial stressor did in negative affect reported in both the self-compassion, B = − 0.19, t
not differ between conditions at any point, ps > .089. See the online (30) = − 5.16, p < .001, and no-strategy control conditions, B = − 0.11, t
supplement for additional details. (27) = − 4.38, p < .001. Following offset of the SR induction, change in
Negative affect. Negative affect across the psychosocial stressor is negative affect did not differ across conditions, B = 0.03, t (57) = 1.34, p
presented by condition in Fig. 2. To investigate the basic pattern of = .184. Exploratory analyses indicated that levels of baseline self-
negative affect across the psychosocial stressor, we first ran a baseline compassion did not moderate the effect of condition on change in
model without any predictors at Level 2. This baseline model indicated negative affect across the psychosocial stressor, ps > .119.
that participants’ average level of negative affect was significantly We then conducted exploratory analyses on each negative affective
different from zero at baseline, B = 1.57, t (58) = 8.87, p < .001, state separately. Negative affect was comprised of participants’ ratings
significantly increased in response to the stressor, B = 0.23, t (58) = of angry, anxious, and upset. At baseline, levels of anger were higher in

Fig. 2. Negative affect ratings for participants in the control and self-compassion conditions. Error bars indicate ± 1 standard error.

5
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

the self-compassion condition than the no-strategy control condition, p promotes effective recovery from stress in participants with elevated
= .034, but changes in anger during the stressor, SR induction, and re­ depressive symptoms.
covery period did not differ between conditions, ps > .132. In addition, Our findings are the first to show that engaging in self-compassion
ratings of anxious and upset affect did not differ at any point, ps > .092. after a stressor improves positive affect – particularly feelings of inspi­
See the online supplement for additional details. ration and happiness – in participants with elevated depressive symp­
toms. These results support correlational research documenting that
3.3. Effect of condition on RSA higher levels of self-compassion are associated with higher levels of
positive affect in daily life (Krieger, Hermann, Zimmermann, & Grosse
RSA across the psychosocial stressor is presented by condition in Holtforth, 2015). To our knowledge, however, this is the first study to
Fig. 3. To examine the basic pattern of RSA response across the stressor, assess levels of positive affect in participants with elevated depressive
a baseline model without any predictors at Level 2 was conducted. This symptoms in response to an experimental self-compassion induction in
model demonstrated that participants average level of RSA at baseline the laboratory. There are far-reaching benefits to increasing positive
differed significantly from zero, B = 6.38, t (54) = 47.04, p < .001. affect in people experiencing depressive symptoms. As many as a third of
Across all participants, levels of RSA then significantly decreased in depressed individuals have clinically significant levels of low positive
response to stress onset, B = − 0.06, t (54) = − 5.78, p < .001, signifi­ affect, or anhedonia (Pelizza & Ferrari, 2009), which has been shown to
cantly increased across the induction period, B = 0.14, t (54) = 6.49, p < predict a poorer course of depression in a number of prospective studies
.001, and finally remained stable across the recovery period, B = 0.03, t (Gentzler & Root, 2019; Kuhlman et al., 2019). There is a longstanding
(54) = 0.62, p = .540. theoretical and empirical literature distinguishing positive and negative
Contrary to expectations, SR condition did not predict RSA levels at affect (Clark & Watson, 1991) and linking them to the approach versus
baseline, RSA reactivity, changes in RSA across the induction period, or withdrawal systems, respectively (Davidson, 2003). Although treatment
RSA recovery, ps > .651. Further, exploratory analyses indicated that for depression has traditionally focused on reducing negative affect,
levels of baseline self-compassion did not moderate the effect of condi­ researchers have increasingly examined the role of targeting positive
tion on change in RSA across the psychosocial stressor, ps > .071. affect (Craske, Meuret, Ritz, Treanor, & Dour, 2016) and findings from
our work suggest that future research might continue to explore whether
4. Discussion self-compassion facilitates changes in both negative and positive affect
domains. Future research might also test whether individual-difference
The current study is the first to experimentally investigate whether factors determine the duration of effectiveness of self-compassion in­
self-compassion was more effective than a control condition at pro­ ductions in the context of depression. Our exploratory analyses indi­
moting recovery from stress in the context of depression as indicated by cated that the self-compassion induction was more effective at
levels of positive affect, negative affect, and RSA. Participants with increasing positive affect for participants’ with higher levels of baseline
elevated depressive symptoms completed a psychosocial stressor and self-compassion. This finding is consistent with other research that
were then randomly assigned to a self-compassion or a no-strategy found inductions were most effective when there was a match between
control condition. We found that, across conditions, participants expe­ induction type and participants’ characteristics (Shull et al., 2016).
rienced the expected affective and physiological responses to the Another important strength of our study is that we assessed re­
stressor. Importantly, we also found that, over and above this within- sponses to stress both during the stress-response induction and 15- and
person change, the SR condition influenced participants’ affective re­ 30-min after induction offset. Past research examining the effects of self-
covery from the stressor. As expected, participants in the self- compassion on responses to stress has focused on the induction period
compassion condition exhibited a greater increase in positive affect exclusively. By assessing the effects of self-compassion post-induction,
and a trend towards a greater decrease in negative affect during the SR we gain additional information about how depressed individuals recover
condition than did participants in the no-strategy control condition. In from stress and the duration over which one single training of self-
contrast, however, changes in RSA levels did not differ between par­ compassion can influence affect. Interestingly, the benefits of the self-
ticipants in the self-compassion and no-strategy control condition. These compassion induction were specific to when participants were
findings provide insight into the degree to which self-compassion engaged in self-compassion and were not observed during the post-

Fig. 3. RSA levels for participants in the control and self-compassion conditions. Error bars indicate ± 1 standard error.

6
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

induction recovery period. One potential reason for the short duration of about careless responding. Another limitation is that our sample was
effects is that the induction was simply too short-lived. A single-session, primarily female, restricting the generalizability of our findings. Future
brief self-compassion instruction may not be sufficient to promote research should recruit a sample with a larger proportion of males in
longer-term changes in affect for people with elevated depressive order to ensure findings generalize across sexes and to test whether sex
symptoms. This possibility can inform future research on self- moderates the findings reported here.
compassion inductions and can encourage future studies to consider The current study is an important addition to the burgeoning liter­
the short-term and long-term dose-response association. It is also ature examining self-compassion as it relates to stress and depression.
important to keep in mind that, during the recovery period, we asked This study extends the research in several ways. It is the first to show
participants to sit quietly without engaging in any activity. We made this that inducing self-compassion supports responses to stressors as
decision in order to examine participants’ natural post-induction re­ measured by positive affect. It is also the first to document that the
covery, which could have been altered by even passive activities benefits of the self-compassion induction on affect were not present 15
(Blagden & Craske, 1996). However, this type of recovery period may or 30 min after induction offset. In addition, evidence that participants
have led to participants’ feeling bored, which can limit experimental in the self-compassion and no-strategy control induction did not differ in
manipulations (Bornstein, Kale, & Cornell, 1990) and may explain why RSA recovery from stress provides insight about the limited benefits of a
levels of positive affect declined during the recovery period. single-session self-compassion induction. These findings expand our
It is also of note that the self-compassion induction was not more understanding of self-compassion and its impact on responses to stress in
effective than the control condition at promoting RSA recovery from the context of elevated depressive symptoms. Given the discrepant re­
stress in our sample of individuals with elevated depressive symptoms. sults between affective and physiological outcomes, further research is
Interestingly, this observation is consistent with evidence that in­ needed to determine whether self-compassion is an effective technique
dividuals high in trait self-compassion do not differ significantly from for promoting effective biological recovery from stressors. It may take
those low in trait self-compassion in several biological markers of stress more time to cultivate and internalize self-compassion. This finding will
recovery, including markers of the autonomic and neuroendocrine sys­ be important to examine further in order to inform future inductions and
tems (Bluth et al., 2016). Research assessing the influence of other experimental work.
emotion regulation strategies on RSA levels have also failed to find
significant effects (Campbell-Sills, Barlow, Brown, & Hofmann, 2006; Funding
Kuo, Fitzpatrick, Metcalfe, & McMain, 2016). For example, Campbell-­
Sills et al. (2006) randomly assigned individuals diagnosed with anxiety This work was supported by the Canadian Institutes of Health
and mood disorders to a single 5-min suppression or acceptance in­ Research [grant number 389703]; the Social Sciences and Humanities
duction, and they found no difference in RSA changes between the two Research Council of Canada [grant number 430-2017-00408]; the Nat­
groups. Thus, it may be more difficult to “move the needle” on biological ural Sciences and Engineering Research Council of Canada [grant
markers of stress via single-session inductions because other factors, number F-17 04841], the University of British Columbia Four Year
such as homeostatic reflexes, influence biological functioning Fellowship For PhD Students #6456, and the Michael Smith Foundation
(Hyndman, 1974). It is also important to consider our results in light of for Health Research. The funding sources were not involved in any stage
the fact that RSA levels are highest during non-stressed or relaxed states. of the research.
In fact, greater effort has long been associated with parasympathetic
withdrawal (Luft, Takase, & Darby, 2009; Lundberg & Frankenhaeuser, Credit author statement
1980), and researchers have documented RSA withdrawal when using
effortful emotion regulation inductions and tasks (LeMoult et al., 2016; Alison Tracy: Conceptualization, Methodology, Investigation,
Reynard, Gevirtz, Berlow, Brown, & Boutelle, 2011). Thus, the effort of Writing – original draft preparation, Project administration. Joelle
applying self-compassion may have influenced RSA levels during the SR LeMoult.: Conceptualization, Methodology, Formal analysis, Writing-
induction. The benefits of self-compassion on RSA may only be observed Reviewing and Editing, Supervision, Funding acquisition. Ellen Jopling:
once self-compassion becomes easier, for example through multiple Formal analysis, Writing – review & editing, Visualization.
training sessions. Supporting this possibility, Arch and colleagues found
that participants who received multiple sessions of self-compassion
training experienced a reduced RSA stress response compared to those Declaration of competing interest
in the no-strategy control condition (Arch et al., 2014).
It is important to acknowledge the limitations of this study and to None.
identify areas for future research. Although participants in the current
study did not meet criteria for major depressive disorder (MDD), levels Appendix A. Supplementary data
of self-compassion scores at baseline (M = 2.4) were consistent with
what other researchers have documented in samples of participants with Supplementary data to this article can be found online at https://doi.
MDD (M = 2.2 in Van Dam, Sheppard, Forsyth, & Earleywine, 2011; M org/10.1016/j.brat.2021.103965.
= 2.74 in Körner et al., 2015). Nonetheless, future research should be
conducted in a clinical sample in order to examine whether the findings
References
reported here are observed in MDD. In addition, a limitation of our study
design is that we did not assess what participants were thinking about American Psychiatric Association. (2013). Diagnostic and statistical manual of mental
following the SR induction period. Assessing participants’ thought disorders (5th). American Psychiatric Association.
Anderl, C., Li, G., & Chen, F. S. (2020). Oral contraceptive use in adolescence predicts
content during the unstructured recovery period might allow us to better
lasting vulnerability to depression in adulthood. Journal of Child Psychology and
understand why the benefits of the self-compassion induction did not Psychiatry, 61(2), 148–156.
continue during the post-induction recovery period. Further, there are Arch, J. J., Brown, K. W., Dean, D. J., Landy, L. N., Brown, K. D., & Laudenslager, M. L.
certain limitations inherent with including university student partici­ (2014). Self-compassion training modulates alpha-amylase, heart rate variability,
and subjective responses to social evaluative threat in women.
pants, particularly when students participate in exchange for course Psychoneuroendocrinology, 42, 49–58.
credit. The majority of participants in the current study received course Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck depression inventory-II.
credit instead of a monetary incentive; thus, they may have been less San Antonio, TX: Psychological Corporation.
Blagden, J. C., & Craske, M. G. (1996). Effects of active and passive rumination and
attentive or careful in their participation. However, all participants distraction: A pilot replication with anxious mood. Journal of Anxiety Disorders, 10
correctly answered attention-check question, attenuating concerns (4), 243–252.

7
A. Tracy et al. Behaviour Research and Therapy 146 (2021) 103965

Bluth, K., Roberson, P. N., Gaylord, S. A., Faurot, K. R., Grewen, K. M., Arzon, S., et al. Luft, C. D. B., Takase, E., & Darby, D. (2009). Heart rate variability and cognitive
(2016). Does self-compassion protect adolescents from stress? Journal of Child and function: Effects of physical effort. Biological Psychology, 82(2), 186–191.
Family Studies, 25(4), 1098–1109. Lundberg, U., & Frankenhaeuser, M. (1980). Pituitary-adrenal and sympathetic-adrenal
Bornstein, R. F., Kale, A. R., & Cornell, K. R. (1990). Boredom as a limiting condition on correlates of distress and effort. Journal of Psychosomatic Research, 24(3–4), 125–130.
the mere exposure effect. Journal of Personality and Social Psychology, 58(5), 791. Maas, C. J., & Hox, J. J. (2005). Sufficient sample sizes for multilevel modeling.
Campbell-Sills, L., Barlow, D. H., Brown, T. A., & Hofmann, S. G. (2006). Effects of Methodology, 1(3), 86–92.
suppression and acceptance on emotional responses of individuals with anxiety and MacBeth, A., & Gumley, A. (2012). Exploring compassion: A meta-analysis of the
mood disorders. Behaviour Research and Therapy, 44(9), 1251–1263. association between self-compassion and psychopathology. Clinical Psychology
Clark, L. A., & Watson, D. (1991). Tripartite model of anxiety and depression: Review, 32(6), 545–552.
Psychometric evidence and taxonomic implications. Journal of Abnormal Psychology, Moncrieff, J., Cohen, D., & Porter, S. (2013). The psychoactive effects of psychiatric
100(3), 316. medication: The elephant in the room. Journal of Psychoactive Drugs, 45(5), 409–415.
Coifman, K. G., & Bonanno, G. A. (2010). When distress does not become depression: Neff, K. D. (2003a). Self-compassion: An alternative conceptualization of a healthy
Emotion context sensitivity and adjustment to bereavement. Journal of Abnormal attitude toward oneself. Self and Identity, 2(2), 85–101.
Psychology, 119(3), 479. Neff, K. D. (2003b). The development and validation of a scale to measure self-
Craske, M. G., Meuret, A. E., Ritz, T., Treanor, M., & Dour, H. J. (2016). Treatment for compassion. Self and Identity, 2(3), 223–250.
anhedonia: A neuroscience driven approach. Depression and Anxiety, 33(10), Neff, K. D. (2016). Self-compassion guided meditations and exercises. Retrieved from
927–938. https://self-compassion.org/category/exercises.
Crowell, S. E., Skidmore, C. R., Rau, H. K., & Williams, P. G. (2013). Psychosocial stress, Neff, K. D., Kirkpatrick, K. L., & Rude, S. S. (2007). Self-compassion and adaptive
emotion regulation, and resilience in adolescence. In Handbook of adolescent health psychological functioning. Journal of Research in Personality, 41(1), 139–154.
psychology (pp. 129–141). New York, NY: Springer. O’Regan, C., Kenny, R. A., Cronin, H., Finucane, C., & Kearney, P. M. (2015).
Davidson, R. J. (2003). Affective neuroscience and psychophysiology: Toward a Antidepressants strongly influence the relationship between depression and heart
synthesis. Psychophysiology, 40(5), 655–665. rate variability: Findings from the Irish longitudinal study on ageing (TILDA).
Diedrich, A., Grant, M., Hofmann, S. G., Hiller, W., & Berking, M. (2014). Self- Psychological Medicine, 45(3), 623–636.
compassion as an emotion regulation strategy in major depressive disorder. Ozer, E. J., Best, S. R., Lipsey, T. L., & Weiss, D. S. (2003). Predictors of posttraumatic
Behaviour Research and Therapy, 58, 43–51. stress disorder and symptoms in adults: A meta-analysis. Psychological Bulletin, 129
Diedrich, A., Hofmann, S. G., Cuijpers, P., & Berking, M. (2016). Self-compassion (1), 52.
enhances the efficacy of explicit cognitive reappraisal as an emotion regulation Pelizza, L., & Ferrari, A. (2009). Anhedonia in schizophrenia and major depression: State
strategy in individuals with major depressive disorder. Behaviour Research and or trait? Annals of General Psychiatry, 8(1), 22.
Therapy, 82, 1–10. Pictet, A., Jermann, F., & Ceschi, G. (2016). When less could be more: Investigating the
Ehret, A. M., Joormann, J., & Berking, M. (2015). Examining risk and resilience factors effects of a brief internet-based imagery cognitive bias modification intervention in
for depression: The role of self-criticism and self-compassion. Cognition and Emotion, depression. Behaviour Research and Therapy, 84, 45–51.
29(8), 1496–1504. Porges, S. W. (2007). The polyvagal perspective. Biological Psychology, 74(2), 116–143.
Ehret, A. M., Joormann, J., & Berking, M. (2018). Self-compassion is more effective than Raes, F., Pommier, E., Neff, K. D., & Van Gucht, D. (2011). Construction and factorial
acceptance and reappraisal in decreasing depressed mood in currently and formerly validation of a short form of the Self-Compassion Scale. Clinical Psychology &
depressed individuals. Journal of Affective Disorders, 226, 220–226. Psychotherapy, 18, 250–255.
Gentzler, A. L., & Root, A. E. (2019). Positive affect regulation in youth: Taking stock and Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data
moving forward. Social Development, 28(2), 323–332. analysis methods (Vol. 1). London: Sage.
Hammen, C. (2005). Stress and depression. Annual Review of Clinical Psychology, 1, Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2004). HLM 6 for windows [computer
293–319. software]. Lincolnwood, IL: Scientific Software International.
Hruschka, D. J., Kohrt, B. A., & Worthman, C. M. (2005). Estimating between-and within- Reynard, A., Gevirtz, R., Berlow, R., Brown, M., & Boutelle, K. (2011). Heart rate
individual variation in cortisol levels using multilevel models. variability as a marker of self-regulation. Applied Psychophysiology and Biofeedback,
Psychoneuroendocrinology, 30(7), 698–714. 36(3), 209.
Hyndman, B. W. (1974). The role of rhythms in homeostasis. Kybernetik, 15(4), 227–236. Rottenberg, J. (2007). Cardiac vagal control in depression: A critical analysis. Biological
Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D. S., Quinn, K., et al. (2010). Psychology, 74(2), 200–211.
Research domain criteria (RDoC): Toward a new classification framework for Rottenberg, J., Wilhelm, F. H., Gross, J. J., & Gotlib, I. H. (2003). Vagal rebound during
research on mental disorders. American Journal of Psychiatry, 167(7), 748–751. resolution of tearful crying among depressed and nondepressed individuals.
Jorm, A. F. (1987). Sex and age differences in depression: A quantitative synthesis of Psychophysiology, 40(1), 1–6.
published research. Australian and New Zealand Journal of Psychiatry, 21(1), 46–53. Scherer, K. R. (2004). Which emotions can be induced by music? What are the
Kessler, R. C., Berglund, P., Demler, O., Jin, R., Koretz, D., Merikangas, K. R., et al. underlying mechanisms? And how can we measure them? Journal of New Music
(2003). The epidemiology of major depressive disorder: Results from the national Research, 33(3), 239–251.
comorbidity survey replication (NCS-R). JAMA, 289(23), 3095–3105. Shull, A., Mayer, S. E., McGinnis, E., Geiss, E., Vargas, I., & Lopez-Duran, N. L. (2016).
Kirschbaum, C., Pirke, K. M., & Hellhammer, D. H. (1993). The ‘Trier Social Stress Trait and state rumination interact to prolong cortisol activation to psychosocial
Test’–a tool for investigating psychobiological stress responses in a laboratory stress in females. Psychoneuroendocrinology, 74, 324–332.
setting. Neuropsychobiology, 28(1–2), 76–81. Soper, D. S. (2020). A-priori sample size calculator for hierarchical multiple regression.
Körner, A., Coroiu, A., Copeland, L., Gomez-Garibello, C., Albani, C., Zenger, M., et al. Software]. Available from: http://www.danielsoper.com/statcalc.
(2015). The role of self-compassion in buffering symptoms of depression in the Speed, M. S., Jefsen, O. H., Børglum, A. D., Speed, D., & Østergaard, S. D. (2019).
general population. PloS One, 10(10), Article e0136598. Investigating the association between body fat and depression via Mendelian
Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological randomization. Translational Psychiatry, 9(1), 1–9.
Psychology, 84(3), 394–421. Tellegen, A., Watson, D., & Clark, L. A. (1999). On the dimensional and hierarchical
Krieger, T., Altenstein, D., Baettig, I., Doerig, N., & Holtforth, M. G. (2013). Self- structure of affect. Psychological Science, 10(4), 297–303.
compassion in depression: Associations with depressive symptoms, rumination, and Tonhajzerova, I., Javorka, M., Trunkvalterova, Z., Chroma, O., Javorkova, J.,
avoidance in depressed outpatients. Behavior Therapy, 44(3), 501–513. Lazarova, Z., et al. (2008). Cardio-respiratory interaction and autonomic dysfunction
Krieger, T., Hermann, H., Zimmermann, J., & Grosse Holtforth, M. (2015). Associations in obesity. Journal of Physiology and Pharmacology, 59(6), 709–718.
of self-compassion and global self-esteem with positive and negative affect and stress Troy, A. S., Wilhelm, F. H., Shallcross, A. J., & Mauss, I. B. (2010). Seeing the silver
reactivity in daily life: Findings from a smart phone study. Personality and Individual lining: Cognitive reappraisal ability moderates the relationship between stress and
Differences, 87, 288–292. depressive symptoms. Emotion, 10(6), 783.
Kuhlman, K. R., Chiang, J. J., Bower, J. E., Irwin, M. R., Cole, S. W., Dahl, R. E., et al. Van Dam, N. T., Sheppard, S. C., Forsyth, J. P., & Earleywine, M. (2011). Self-compassion
(2019). Persistent low positive affect and sleep disturbance across adolescence is a better predictor than mindfulness of symptom severity and quality of life in
moderate link between stress and depressive symptoms in early adulthood. Journal of mixed anxiety and depression. Journal of Anxiety Disorders, 25(1), 123–130.
Abnormal Child Psychology, 1–13. Voss, A., Schroeder, R., Heitmann, A., Peters, A., & Perz, S. (2015). Short-term heart rate
Kuo, J. R., Fitzpatrick, S., Metcalfe, R. K., & McMain, S. (2016). A multi-method variability—influence of gender and age in healthy subjects. PloS One, 10(3), Article
laboratory investigation of emotional reactivity and emotion regulation abilities in e0118308.
borderline personality disorder. Journal of Behavior Therapy and Experimental Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief
Psychiatry, 50, 52–60. measures of positive and negative affect: The PANAS scales. Journal of Personality
Lee, J. S., Mathews, A., Shergill, S., & Yiend, J. (2016). Magnitude of negative and Social Psychology, 54(6), 1063.
interpretation bias depends on severity of depression. Behaviour Research and Waugh, C. E., Muhtadie, L., Thompson, R. J., Joormann, J., & Gotlib, I. H. (2012).
Therapy, 83, 26–34. Affective and physiological responses to stress in girls at elevated risk for depression.
LeMoult, J., Yoon, K. L., & Joormann, J. (2016). Rumination and cognitive distraction in Development and Psychopathology, 24(2), 661–675.
major depressive disorder: An examination of respiratory sinus arrhythmia. Journal
of Psychopathology and Behavioral Assessment, 38(1), 20–29.