You are on page 1of 11

Cogn Ther Res (2014) 38:189–199

DOI 10.1007/s10608-013-9581-9

ORIGINAL ARTICLE

Rumination Moderates the Effects of Cognitive Bias Modification


of Attention
Kimberly A. Arditte • Jutta Joormann

Published online: 29 September 2013


Ó Springer Science+Business Media New York 2013

Abstract Whereas research on cognitive bias modifica- in the etiology and maintenance of emotional disorders
tion of attention has produced promising results, it remains (e.g., Beck and Clark 1997; Beck et al. 1979). Specifically,
unclear how and for whom such techniques may be most it has been proposed that cognitive biases are related to
effectively implemented. This study examined how trait impaired ability to regulate affect, thereby affecting stress
rumination moderated the effects of two attention training responding and recovery. These difficulties in affect reg-
tasks on biased attention, assessed via eye tracking, and ulation result in the sustained negative affect and difficul-
subsequent stress reactivity. Using a modified dot-probe ties experiencing positive affect that characterize mood
task, participants were trained to develop a negative disorders (e.g., Joormann 2010). Research has supported
attention bias, a positive attention bias, or no attention bias. theory on attention and emotion, finding that an attentional
Though neither the train-negative nor the train-positive preference for negative stimuli (negative attention bias
conditions produced significant main effects on attention [NAB]) is related to heightened emotional reactivity (El-
biases or emotional reactivity, rumination was found to lenbogen et al. 2002), as well as clinical symptoms of
moderate training efficacy, such that train-positive partic- depression and anxiety (for a meta-analytic review, see
ipants reporting high levels of rumination demonstrated Armstrong and Olatunji 2012). Conversely, an attentional
greater early-stage positive attention biases at post-training, preference for positive stimuli (positive attention bias
as compared to controls. Further, train-negative and train- [PAB]) appears to be associated with greater positive
positive participants who reported low levels of rumination affect, lesser negative affect, and reduced levels of physi-
demonstrated greater positive affect following an acute ological arousal in response to stress (Segerstrom 2001;
stressor as compared to controls. Theoretical and clinical Strauss and Allen 2006).
implications of results are discussed. Despite all that is known about attention biases, much of
the extant research has been limited by its correlational
Keywords Cognitive bias modification  Attention nature. As a result, it remains unclear whether biased atten-
bias modification  Eye tracking  Attention bias  tion represents a causal mechanism that leads to altered
Emotional reactivity emotional responding or merely a state-dependent symptom
of such responding. To address the causal impact of biased
attention on emotion, research has begun to experimentally
Introduction manipulate biases. Cognitive bias modification of attention
(CBM-A) has the potential to increase our understanding of
The preference, or bias, to attend to emotion-congruent the interaction of cognition and emotion and is also thought
environmental stimuli is posited to play an important role to be an important next step in translating basic research into
clinically-relevant interventions. Initial findings have pro-
duced promising, albeit modest, effects. CBM-A tasks have
K. A. Arditte (&)  J. Joormann
been shown to effectively alter participants’ ensuing emo-
Department of Psychology, University of Miami, 5665 Ponce de
Leon Boulevard, Coral Gables, FL 33146, USA tional responsivity and reduce clinical symptoms, particu-
e-mail: karditte@psy.miami.edu larly symptoms of anxiety (for reviews, see Beard et al. 2012;

123
190 Cogn Ther Res (2014) 38:189–199

Hallion and Ruscio 2011). Few studies, however, have CBM-A moderation, as attentional deployment is, itself,
assessed the impact of CBM-A on symptoms of depression, conceptualized as an emotion regulation strategy that is
and those that have, have produced equivocal results (e.g., commonly used to up- or down-regulate one’s emotional
Wells and Beevers 2010, but Baert et al. 2010). Thus, response and which has been found to relate to the use of
important questions about the utility of CBM-A remain other emotion regulation strategies (Gross and Thompson
unanswered. 2007; Wadlinger and Isaacowitz 2011). Thus, reliance on
For one, the literature lacks a complete understanding of specific emotion regulation strategies that are associated
how CBM-A paradigms affect change. Altered attention is with attentional deployment may be expected to impact
often presumed to be the mechanism of change underlying training efficacy.
CBM-A, but other factors (e.g., increased self-regulation It has been theorized that rumination, an emotion reg-
and effortful control; Rueda et al. 2005) may also be at ulation strategy that involves perseverative, introspective
play. Whereas the impact of CBM-A on attention for focus on negative thoughts and feelings, is associated with
emotional stimuli is undoubtedly a central focus of the attentional deployment, such that individuals who rely on
literature, training studies have not always included a this strategy demonstrate difficulty disengaging with both
systematic, pre-post assessment of attention. Further, when negative thoughts and negative environmental stimuli
they have, assessments have typically been limited in (Johnson 2009). Research suggests that rumination is
important ways. associated with attentional inflexibility, driven by difficulty
For example, most studies have assessed change in inhibiting response to negative emotional material and
attention biases using the dot-probe task (e.g., Amir et al. impairment in set-shifting ability (De Lissnyder et al.
2008; Baert et al. 2010; MacLeod et al. 2002; Wells and 2010). Further, research has linked rumination with biased
Beevers 2010). Yet, this task is limited in that it can only attention, finding that it is uniquely associated with
provide a cross-sectional snapshot of attention allocation at increased attention for sad stimuli in both children and
the time the probe appears and can miss key information on adults (Joormann et al. 2006; Romens and Pollack 2012).
how attention to emotional stimuli changes over time In an initial attempt to link CBM-A with the habitual use of
(Caseras et al. 2007). In contrast to the dot-probe, eye rumination, Baert et al. 2010 assessed change in self-
tracking can provide a more ecologically valid under- reported rumination as a function of training. Whereas the
standing of biased attention because it allows for the con- authors found a general decline in rumination scores from
tinuous evaluation of both early- and late-stage focal pre- to post-training, there was no difference as a function
attention. A number of previous CBM-A studies have also of training condition. Rather than examining the impact of
been limited by their use of the same stimuli within both a brief attention training on habitual use of this strategy,
the assessment and training phases (Baert et al. 2010) or by however, it may be more helpful to explore the moderating
their use of some component of the training phase as an effect of this trait-level characteristic on the effects of
assessment (MacLeod et al. 2002; Wells and Beevers CBM-A.
2010). Yet, to fully understand the implications of CBM-A, The current investigation utilized an attention training
it will be important to examine whether training effects paradigm to alter attention towards positive or negative
generalize to other content-similar stimuli (MacLeod and stimuli (referred to as the PAB and NAB conditions,
Mathews 2012). respectively). These conditions were compared to a no-
Moreover, with the understanding that on the whole training control (CTL) condition in order to assess the
CBM-A has produced only modest effects on subsequent causal impact of attention biases on natural gaze patterns
emotional responding (e.g., Beard et al. 2012; Hallion and and subsequent stress reactivity. The study also sought to
Ruscio 2011), it will be essential to begin to identify who understand how individual differences in habitual use of
stands to benefit from attention training. Aside from rumination would moderate the effects of the training on
comparing clinical and non-clinical populations, CBM-A attention biases and emotional responding. It was hypoth-
research must explore potential moderators of training esized that the completion of the attention training would
effects, such as trait-level differences in the use of specific alter attentional patterns assessed via eye tracking before
emotion regulation strategies within both clinical and non- and after training. Specifically, it was predicted that,
clinical populations. Taking an individual differences compared to controls, participants in the NAB condition
approach to the study of attention biases may provide would demonstrate increased attention for negative stimuli
insight into the mechanisms by which CBM-A elicits at post-training, whereas participants in the PAB condition
change in emotional responding and may aid us in targeting would demonstrate increased attention to positive stimuli at
individuals who stand to benefit from completion of CBM- post-training. It was also hypothesized that completion of
A procedures. Examining the use of specific emotion reg- the attention training would alter patterns of emotional
ulation strategies is a natural starting point for research on reactivity in response to a laboratory stressor. Here it was

123
Cogn Ther Res (2014) 38:189–199 191

predicted that, as compared to the CTL condition, partici- emotion ‘‘right now, in this moment.’’ The psychometric
pants completing the NAB training would demonstrate properties of these scales have been assessed in a sample of
greater emotional reactivity (i.e., more subjective negative undergraduate students. Results indicate that the measure is
affect/less subjective positive affect), whereas participants internally consistent (a’s ranging from .86 to .90), has
completing the PAB training would demonstrate less strong convergent and divergent validity, and is reliable
emotional reactivity (i.e., less subjective negative affect/ across time (Watson et al. 1988).
more subjective positive affect). Finally, it was predicted
that habitual use of rumination would moderate training, Attention Training Task
though given the exploratory nature of this hypothesis, no
specific predictions were made about the direction of such Task
moderation effects.
Participants were randomly assigned to one of three
attention training conditions: NAB, PAB, or CTL. Across
Method conditions, training tasks consisted of 10 practice trials and
160 experiment trials. Participants began each trial by
Participants fixating their gaze on a black cross presented in the middle
of the screen (500 ms). Then, two faces appeared simul-
A sample of N = 104 undergraduate students was recruited taneously on either side of the screen (500 ms). Finally, a
for the current study. Eligible participants were required to probe (the letter E or the letter F) appeared on either the left
be at least 18 years of age or have parental consent and to or right side of the screen, in the spatial location of one of
be currently enrolled as a student at the University of the two faces. Participants were instructed to identify the
Miami. probe as quickly as possible. Probe detection accuracy and
response latencies were recorded.
Self-Report Measures Participants assigned to the NAB condition were pre-
sented with 128 (80 %) negative-neutral trials and 32
Rumination (20 %) neutral–neutral trials. Similarly, participants
assigned to the PAB condition were presented with 128
The Ruminative Responses Scale (RRS; Treynor et al. (80 %) positive-neutral trials and 32 (20 %) neutral–neu-
2003) is a 22-item questionnaire that assesses the habitual tral trials. Image location (i.e., valenced image located on
use of rumination in response to dysphoric mood. In par- the left or right side of the screen) and probe type (E or F)
ticular, the measure examines responses that are focused on were counterbalanced across trials. In both the NAB and
the self, on symptoms, or on possible consequences and the PAB conditions, the probe appeared behind the val-
causes of moods using a 4-point scale (Almost never to enced image in each of the 128 valenced-neutral trials.
Almost always). The RRS has two subscales, brooding Thus, as participants completed the training it was expec-
(e.g., ‘‘I think ‘what am I doing to deserve this’’) and ted that they would learn to shift their attention towards the
reflection (e.g., ‘‘Analyze recent events to try to understand valenced image. Previous research has found similar par-
why you are depressed’’). Subscales are then summed to adigms to effectively induce negative and positive attention
create a total score that ranges from 22 to 88, with higher biases (e.g., Amir et al. 2008).
scores indicating greater habitual reliance on rumination. In the CTL condition, participants were presented with
Internal consistency (a = .76 across subscales) and test– 64 (40 %) negative-neutral, 64 (40 %) positive-neutral, and
retest reliability (r = .61 across subscales) have been pre- 32 (20 %) neutral–neutral images. As with the NAB and
viously determined within a community sample (Treynor PAB conditions, image location and probe type were
et al. 2003). counterbalanced. Additionally, within the CTL condition,
probe location was counterbalanced, such that the probe
Subjective Mood Ratings could appear behind either the valenced or the neutral
image. It was expected that participants in the CTL con-
Participants’ self-reported affect was monitored throughout dition would not learn to shift attention towards valenced
the study using the Positive Affect Negative Affect Scales stimuli and, thus, would not develop an attention bias.
(PANAS; Watson et al. 1988). The PANAS consists of
two, 10-item mood scales: positive (e.g., ‘‘enthusiastic’’) Stimuli
and negative (e.g., ‘‘upset’’) affect. Participants are asked
to rate each emotion on a scale of 1 (Not at all) to 5 Images used during the attention training were selected
(Extremely) based on how much they are experiencing that from the NimStim Face Stimulus Set (MacArthur

123
192 Cogn Ther Res (2014) 38:189–199

Foundation Research Network on Early Experience and neutral images (M = 4.99, SD = .34), t (97) = 28.28,
Brain Development).1 Participants were presented with 10 p \ .001, d = 5.74, and neutral images rated as less
expression pairs (i.e., neutral–neutral expressions from the pleasurable than positive images (M = 7.45, SD = .31),
same model) during the practice phase of the task and 16 t (97) = 29.37, p \ .001, d = 5.96. The categories also
expression pairs (i.e., sad-neutral, happy-neutral, or neu- differed significantly in their arousal ratings, F(2, 116) =
tral–neutral expressions from the same model) during the 103.02, p \ .001, g2 = .64. Negative (M = 4.94, SD =
training phase of the task. Face stimuli were selected to .45) and positive (M = 4.87, SD = .69) images were sig-
control for the gender and race of the model. In addition, nificantly more arousing than the neutral images (M =
the original stimuli were edited so that only the facial 3.17, SD = .64), t (97) = 11.58, p \ .001, d = 2.35, and
features of each model were visible to the participant. t (97) = 10.44, p \ .001, d = 2.12 for negative-neutral
and positive-neutral comparisons, respectively, but not
Assessment of Attention Biases significantly different from each other, t (38) = .34,
p = .74, d = .11.
Task
Eye Tracking
A passive viewing paradigm was used pre- and post-
attention training to assess change in attention to positive During the passive viewing paradigm, bilateral eye
and negative stimuli. During this task, participants viewed movements were unobtrusively tracked using a Tobii X120
a series of 60 trials, each of which included two images Eye Tracker and Tobii Studio Analysis Software (Tobii
selected from the International Affective Picture System Technology, Danderyd, Sweden). Eye tracking data were
(IAPS; NIMH Center for the Study of Emotion and recorded at a rate of 60 times per second (60 Hz) during the
Attention 1999). Trials were presented in six blocks of ten, 6,000 ms per trial when the IAPS images were presented.
categorized by valence. Block order and image location Data on four indices of attention were collected for each
were randomized to control for order effects. Each trial of the prescribed areas-of-interest (i.e., each of the two
began with the presentation of a black fixation cross in the images within a given trial). Early-stage attention biases
middle of the screen (500 ms), followed by a number from were assessed by initial orientation (IO), or the decision to
one to five (500 ms). To ensure that participants began initially attend to one image over another. IO was calcu-
each trial with eye gaze directed at the center of the screen, lated by examining the time to first fixation, with fixation
participants were required to say this number aloud. Trials defined as the maintenance of gaze on a single location, for
ended with the presentation of a picture pair. The images both the valenced and neutral images presented in a given
remained on the screen for 6,000 ms, during which, par- trial. As with previous research (e.g., Caseras et al. 2007;
ticipants were instructed to look at whatever they liked. Armstrong et al. 2010), bias scores were calculated by
summing the number of trials in which the participant first
Stimuli oriented to the valenced image and dividing this by the
total number of valid trials. Bias scores were calculated
Attention for positive and negative stimuli was assessed separately for negative-neutral and positive-neutral trials
using 20 positive-neutral and 20 negative-neutral (negative with a score of .50 (i.e., 50 %) indicating no bias, a score of
images were dysphoric in nature) picture pairs, respec- [.50 indicating bias towards emotional images, and a score
tively. In addition, 20 neutral–neutral picture pairs were \.50 indicating bias away from emotional images.
included in the task. In keeping with the study’s attention Each of the remaining three indices was calculated as an
training tasks, neutral trials were included as filler trials indicator of attentional maintenance. Fixation count (FC)
and were intended to mask the purpose of the passive was assessed by examining the number of times a partici-
viewing task. IAPS image pairs were selected for use in pant fixated on each of the images in a given trial. In
this task based on normed valence and arousal ratings contrast, visit count (VC) was assessed by examining the
(Lang et al. 2005) and previous research (e.g., Caseras number of visits to each image in a given trial. A visit was
et al. 2007; See Footnote 1). The three categories of images defined as the time spent looking at an image without
differed significantly in their valence ratings, F(2, disengaging from it, such that one visit could contain an
116) = 1,013.97, p \ .001, g2 = .95, with negative ima- indefinite number of fixations. VC is, thus, conceptualized
ges (M = 2.44, SD = .43) rated as less pleasurable than as the number of times a participant engaged, disengaged
and then re-engaged with an image over the course of a
1 trial. Finally, total visit duration (TVD) was assessed by
For information on specific stimuli used in the attention training
task and/or in the assessment of attention biases, please contact the examining the length of time a participant spent examining
first author, Kimberly A. Arditte, at karditte@psy.miami.edu. each image over the course of a trial. Bias scores for each

123
Cogn Ther Res (2014) 38:189–199 193

of these indices were calculated by finding the mean for speech preparation period, after which a third and final
valenced images and subtracting from this the mean for mood rating was collected. Once the participants had
neutral images (e.g., Armstrong et al. 2010). As with IO, provided this mood rating, the experimenter explained to
separate scores were calculated for negative-neutral and participants that ‘‘only about half of the people enrolled in
positive-neutral trials. With the indices of attentional the study’’ would actually have to give the speech. Whether
maintenance, a score of zero represented no bias, a positive or not a participant had to give a speech was ‘‘determined’’
value indicated bias towards emotional images, and a by the flip of a coin. The experimenter then flipped a
negative value indicated bias away from emotional images. quarter (which, unbeknownst to participants was double-
sided) and ‘‘determined’’ that the participant would not
Assessment of Emotional Reactivity have to give the speech. Participants were debriefed on the
use of this minor deception, as well as the study’s aims and
To examine the effects of the attention training on emo- hypotheses prior to the conclusion of study sessions.
tional reactivity, participants underwent a speech prepara-
tion task designed to elicit stress and anxiety (Sanchez
et al. 2012). Participants were instructed that they would Results
have to prepare and deliver a 5-min speech on the topic
‘‘Why am I a good friend?’’ To increase the stressfulness of Following randomization, n = 35 participants were
the task participants were informed that their speech would assigned to the NAB condition, n = 34 were assigned to
be video recorded and judged based on factors such as the PAB condition, and n = 35 were assigned to the CTL
clarity, cohesiveness, and persuasiveness. In addition, they condition. Data on age were missing from four participants
were told that their speech could be used for training and complete data from the RRS were missing from eight
purposes at the university, with the top 10 % and the bot- participants. Throughout analyses, we have opted to use all
tom 10 % of speeches based on the evaluators rating being available and valid data. On average, participants were
used as examples of strong and weak public speaking 19.68 (SD = 2.27) years. The sample was 53 % female
skills, respectively. After receiving these instructions, and 47 % male. Further, 56 % of the sample identified as
participants were given 5 min to prepare their speech and White/Caucasian, 15 % as Hispanic/Latino, 12 % as Black/
were allowed a pen and paper to make notes. African American, 8 % as Asian, and 9 % as ‘‘other.’’
Given the breakdown of participants’ reported race/eth-
Procedure nicity this variable was dichotomized as White/Caucasian
versus non-White/Caucasian for subsequent analyses.
The study protocol was reviewed and approved by the Demographic characteristics within each training condi-
University of Miami’s Institutional Review Board. Prior to tion are presented in Table 1. Results revealed that random
providing consent, participants were told that the purpose assignment of participants to attention training condition was
of the study was to investigate how people respond to effective at minimizing between-group differences on
positive and negative information. Participants were not demographic variables of interest. Study conditions did not
provided with explicit information about study aims or differ as a function of age, F(2, 97) = .18, p = .84,
hypotheses, and they were not informed that they would g2 = .003, gender, v2 (2, N = 104) = 3.30, p = .19, or
complete an attention training task. After providing con- race/ethnicity v2 (2, N = 104) = 1.75, p = .42. The aver-
sent, participants completed the RRS and a questionnaire age score on the RRS was 39.23 (SD = 11.70). Variability in
assessing relevant demographic information. Upon com- RRS scores appeared to be evenly distributed across training
pletion of these measures, participants watched a brief (5- tasks and no significant differences among conditions were
min) nature video. They were, then, asked to provide a found, F(2, 93) = .90, p = .41, g2 = .02.
mood rating, which was used as a measure of baseline Attention training performance was examined by con-
positive and negative affect. Next, participants underwent ducting one-way ANOVAs to examine between-group
the pre-training assessment of attention, before being differences on mean accuracy and reaction time as a
assigned to complete one of the three attention training function of training condition. Overall, participants
tasks. At the conclusion of the attention training, partici- were highly accurate in their response to task probes
pants provided a second mood rating. Participants, then, (M = 97 %, SD = 3 %) and no difference in accuracy was
underwent the second assessment of attention. The proce- found as a function of training condition, F(2, 101) = 2.38,
dure for this portion of the protocol was identical to that of p = .10, g2 = .04. Training conditions also did not differ
the pre-training assessment. in their mean reaction times (RT) to trial probes, F(2,
Finally, the experimenter provided the instructions for 101) = 1.00, p = .37, g2 = .02. Across conditions, par-
the stressor task. Instructions were followed by a 5-min ticipants were relatively quick to respond (M = 773.20 ms,

123
194 Cogn Ther Res (2014) 38:189–199

Table 1 Means (standard deviations) and percentages (where speci- Table 2 Means (standard deviations) for pre-training and post-
fied) for demographic variables of interest and habitual use of rumi- training eye tracking indices, including initial orientation (IO), fixa-
nation, as measured by the Ruminative Responses Scale (RRS), tion count (FC), visit count (VC), and total visit duration (TVD) for
across the negative attention bias (NAB), positive attention bias negative (Neg) and positive (Pos) stimuli, across the negative atten-
(PAB), and control conditions tion bias (NAB), positive attention bias (PAB), and control conditions
NAB PAB Control NAB PAB Control

Age 19.50 (2.66) 19.71 (2.08) 19.83 (2.09) Pre-training


Gender (% female) 62.80 41.17 54.28 IO-Neg .52 (.07) .53 (.06) .51 (.05)
Race/ethnicity (% 34.28 47.05 48.57 IO-Pos .61 (.09) .61 (.09) .59 (.07)
White/Caucasian) FC-Neg 1.93 (1.49) 1.06 (1.56) 1.58 (1.56)
RRS 37.22 (12.49) 39.38 (10.93) 41.16 (11.63) FC-Pos 1.77 (1.57) 1.87 (1.29) 1.82 (1.47)
VC-Neg .08 (.18) .00 (.20) .13 (.22)
VC-Pos .17 (.27) .15 (.16) .15 (.22)
SD = 90.79 ms) and no individual participant had a mean TVD-Neg .68 (.53) .45 (.61) .69 (.79)
RT above 1,250 ms. To examine if conditions did, indeed, TVD-Pos .63 (.57) .62 (.53) .77 (.73)
train attention for positive and negative stimuli, we com- Post-training
pared RTs in the first and second halves of the training IO-Neg .52 (.10) .52 (.15) .53 (.06)
within each of the three training conditions. Results indi- IO-Pos .57 (.10) .55 (.15) .58 (.09)
cated no difference in RTs when collapsing across all FC-Neg 1.66 (1.85) 1.03 (1.29) .86 (2.14)
training trials, F(2, 101) = 1.06, p = .35, g2 = .02, or FC-Pos 1.84 (1.39) 2.17 (1.95) 1.78 (1.26)
when assessing condition-congruent trials, F(2, 101) =
VC-Neg .05 (.23) .05 (.20) .00 (.26)
.96, p = .38, g2 = .02.
VC-Pos .14 (.14) .14 (.22) .15 (.18)
Following completion of data collection, but prior to
TVD-Neg .54 (.74) .37 (.62) .16 (.95)
analysis, eye tracking data were examined for outliers.
TVD-Pos .65 (.58) .65 (.67) .79 (.87)
Participants’ data were included in analyses if valid data
were collected for at least 70 % of the stimulus presenta-
tion duration collapsed across trials. Of the 104 partici-
pants, valid data were collected from n = 72 participants at Next, habitual use of rumination was explored as a
both the pre-training and post-training assessments (n = 26 moderator of training effects using multiple regression
NAB participants, n = 24 PAB participants, and n = 22 analyses. To do this, the condition variable was dummy
CTL participants). For an additional four cases (one NAB coded, such that the control condition served as the referent
and three PAB participants), the IO index could not be group. Participants’ RRS scores were, then, centered to
calculated due to insufficient data. For the subset of par- examine the main effect of condition at mean levels of
ticipants included in eye tracking analyses, the average rumination. Finally, regressions of eye tracking indices on
percentage of data collected was 89.64 (SD = 6.09) during rumination were specified separately for each condition
the pre-training assessment and 86.38 (SD = 7.01) during (i.e., the model included interaction terms for rumination
the post-training assessment. by condition). Condition dummy codes, the centered RRS
Eye tracking data were first examined using a series of term, and Condition 9 RRS interaction terms were
mixed-factor ANOVAs that included Time (pre- and post- simultaneously entered into a series of regression models
training) as a within-subjects variable and Condition examining post-training attention biases. In all cases, the
(NAB, PAB, or CTL) as a between-subjects variable (see bias score for the relevant pre-training index of attention
Table 2 for descriptive statistics). Results revealed no was also included as a predictor to control for baseline
significant Time 9 Condition interactions for the four biases across participants. The approach of controlling for
indices of attention for negative stimuli: IO, F(2, 65) = baseline biases by entering pre-training scores as a covar-
.26, p = .77, partial g2 = .01; FC, F(2, 67) = .96, iate in our regression model was selected over the use of a
p = .39, partial g2 = .03; VC, F(2, 67) = 2.32, p = .11, pre-post difference scores, because differences scores may
partial g2 = .07; TVD, F(2, 67) = 2.30, p = .11, partial be systematically related to random error and are often
g2 = .06. Likewise, no significant Time 9 Condition correlated with baseline scores (Cronbach and Furby 1970;
interactions emerged from analyses examining attention for Cohen et al. 2003). Regressions were used to separately
positive stimuli: IO, F(2, 65) = .74, p = .48, partial examine attention for positive and negative stimuli for each
g2 = .02; FC, F(2, 67) = .19, p = .83 partial g2 = .01; of the four eye tracking indices (IO, FC, VC, and TVD).
VC, F(2, 67) = .16, p = .86, partial g2 = .01; TVD, F(2, A significant interaction emerged from regression
67) = .001, p = .99, partial g2 \ .001. analyses, which demonstrated that rumination impacted the

123
Cogn Ther Res (2014) 38:189–199 195

effects of the PAB, versus CTL, condition on participants’ training, the relation remained non-significant in the CTL
IO for positive stimuli (b = .41, t [55] = 2.19, p = .03; condition (r = -.24, p = .30), but was significant within
see Fig. 1). Methods for probing this interaction were the PAB condition (r = .45, p = .04). As rumination
based on those laid out by Preacher et al. (2004) and increased, so too did PAB participants’ early-stage positive
involved examining the simple slopes, as well as the simple attention biases.
intercepts, of the model. Simple slopes were examined by None of the other regression analyses yielded evidence
separately regressing post-training IO on RRS scores for for rumination as a moderator of training effects on eye
the PAB and CTL conditions. For participants within the tracking indices (all p’s [ .10). No other interaction effects
PAB condition, rumination predicted greater post-training emerged from analyses. Contrary to hypotheses, rumina-
bias scores (b = .59, t [55] = 2.01, p = .049). In contrast, tion did not interact with the NAB training to alter early-
the relation between rumination and post-training biases (IO: b = -.32, t [55] = 1.29, p = .20) or late-stage (FC:
was not significant for participants assigned to the control b = .12, t [57] = .46, p = 65; VC: b = .08, t [57] = .32,
condition (b = -.32, t [55] = 1.04, p = .30). The simple p = .75; TVD: b = .22, t [57] = 1.59, p = .12) attention
intercepts of the model were examined, by regressing post- for negative stimuli. Rumination also did not moderate the
training IO on the condition variable at different levels of effects of the PAB training on late-stage attention for
the moderator (e.g., ±1 SD from the mean; Preacher et al. positive stimuli, as assessed by FC (b = .02, t [57] = .11,
2004). Here, the b value represents the vertical difference p = .94), VC (b = .28, t [57] = 1.48, p = .15), or TVD
between the two regression lines. It is interpreted as the (b = .01, t [57] = .08, p = .94).
estimated mean difference at each of the three levels of the As with analyses examining attention, data on emotional
moderator and is tested for statistical significance using a reactivity were first examined with a series of mixed-factor
t test (see Eckenrode et al. 2000). Testing the simple ANOVAs. Again, Time (mood ratings following baseline,
intercepts, it was found that participants reporting high training, and stressor task periods) was included as a
levels of habitual rumination and who were assigned to the within-subjects factor and Condition as a between subjects
PAB condition demonstrated more pronounced biases at factor, with rumination excluded from analyses (see
post-training, compared to their counterparts in the control Table 3 for descriptive statistics). As expected, a main
condition (b = .52, t [55] = 2.03, p = .047). Yet, partic- effect of time was found for negative affect, F(2,
ipants reporting low levels of rumination, did not differ in 196) = 51.95, p \ .001, partial g2 = .35; across condi-
post-training biases as a function of assignment to the PAB tions, participants’ negative affect increased from baseline
or control condition (b = .38, t [55] = 1.46, p = .15). To (M = 11.34, SD = 2.28) to training (M = 12.37, SD =
further support regression analyses, the relations between 3.03), t (101) = 3.40, p = .001, d = 1.37, and from
pre- and post-training IO for positive stimuli and RRS training to stressor (M = 15.58, SD = 5.09), t (101) =
scores were examined using zero-order correlations. The 6.59, p \ .001, d = 1.54. Regardless of training condition,
relation between pre-training IO for positive stimuli and the stressor task effectively elicited negative affect. A main
RRS scores were non-significant for the CTL (r = .35, effect of time was also found for positive affect, F(2,
p = .30) and PAB (r = .24, p = .30) conditions. At post- 200) = 30.89, p \ .001, partial g2 = .40. Across condi-
tions, positive affect decreased from baseline (M =
0.8 19.29, SD = 6.76) to training (M = 15.27, SD = 5.61),
PAB
0.75 Control
Table 3 Means (standard deviations) for subjective negative and
positive affect at baseline, as well as immediately after the attention
0.7
training and stressor tasks, across the negative attention bias (NAB),
Bias Score

* positive attention bias (PAB), and control conditions


0.65
NAB PAB Control
0.6
Baseline
0.55 Negative affect 12.09 (3.21) 11.03 (1.74) 10.97 (1.45)
0.5
Positive affect 20.41 (6.43) 18.53 (6.82) 19.11 (7.08)
-1 SD +1 SD Post-training
RRS Score Negative affect 11.85 (2.56) 12.82 (3.65) 12.51 (2.82)
Positive affect 15.50 (5.44) 15.06 (4.68) 15.23 (6.73)
Fig. 1 Ruminative Responses Scale (RRS) score 9 Condition (posi-
tive attention bias [PAB] versus Control) interaction effects on early- Post-stressor
stage attention biases (i.e., initial orientation) for positive stimuli at Negative affect 15.70 (4.96) 15.33 (4.57) 15.69 (5.82)
post-training; solid line indicates p \ .05; * = p \ .05; all other Positive affect 20.74 (8.03) 20.09 (8.98) 22.97 (8.50)
comparisons, p [ .10

123
196 Cogn Ther Res (2014) 38:189–199

t (103) = 6.89, p \ .001, d = 3.50, but increased again 25


following the stressor (M = 21.28, SD = 8.52), NAB
t (102) = 6.85, p \ .001, d = 2.09. Despite these main PAB
effects, neither negative affect, F(4, 196) = * Control

Positive Affect
2 *
1.18, p = .32, partial g = .02, nor positive affect, F(4,
200) = 1.11, p = .35, partial g2 = .02, differed as a func- 20
tion of condition across time.
Multiple regression models to examine the interactive
effects of rumination and training condition on positive and
negative affect were constructed in a manner consistent
with attention analyses. Dummy coded condition variables, 15
the centered RRS term, and Condition 9 RRS interaction -1 SD +1 SD
RRS Score
terms were simultaneously entered into a model, which
also controlled for baseline affect, to predict subjective Fig. 2 Ruminative Responses Scale (RRS) score 9 Condition (neg-
affect following the stressor. Separate analyses were con- ative attention bias [NAB] vesus control; positive attention bias
ducted for positive and negative affect. Rumination did not [PAB] versus control) interaction effects on subjective positive affect
following completion of the stressor task; * = p \ .05; all other
appear to moderate the attention training effects on nega- comparisons, p [ .10
tive affect (p’s [ .10). However, after controlling for
baseline levels of affect, nearly significant effects emerged
for interaction terms comparing participants assigned to the support regression analyses, the relations between baseline
NAB versus control condition (b = -.27, t [86] = 1.97, and stressor positive affect and RRS scores were examined
p = .05) and PAB versus control condition (b = -25, using zero-order correlations. Though no significant cor-
t [86] = 1.93, p = .06), predicting positive affect follow- relations emerged from these analyses, the relation between
ing the stressor (see Fig. 2). positive affect following the stressor and RRS scores was
Once again, interactions were probed by examining both marginally significant (r = -.30, p = .09).
simple slopes and simple intercepts. Compared to partici-
pants in the control condition, individuals assigned to the
NAB (b = .30, t [86] = 2.18, p = .03) or PAB (b = .30, Discussion
t [86] = 2.16, p = .03) conditions and who reported low
levels of rumination demonstrated greater positive affect The current study examined the effects of two CBM-A
following the stressor. However, among participants tasks on attention biases and subsequent emotional reac-
endorsing high levels of habitual rumination, conditions tivity. It was predicted that, compared to controls, partici-
did not differ in self-reported positive affect (p’s [ .10). pants who were trained to attend to negative stimuli would
Moreover, while none of the simple slopes were found to demonstrate more pronounced NABs and greater sub-
be statistically significant (p’s [ .10), the direction of these sequent stress reactivity. Conversely, participants trained to
effects was, perhaps, notable. Among participants assigned attend to positive stimuli were expected to demonstrate
to the CTL condition, rumination was positively related to greater PABs and attenuated subsequent stress reactivity as
positive affect following the stressor. In contrast, partici- compared to controls. Moreover, it was expected that
pants assigned to both the NAB and PAB conditions habitual use of rumination would moderate training effects.
demonstrated inverse relations between rumination and Excluding rumination from analyses, neither the NAB
subsequent positive affect (See Footnote 2).2 To further nor the PAB condition produced significant effects on

Footnote 2 continued
2
Though a systematic examination of the moderating effects of depression was significantly associated with negative affect at base-
depression and social anxiety symptoms were beyond the scope of line (r = .31, p = .002), whereas social anxiety was related to neg-
this paper, analyses were conducted to ensure that the reported ative affect following the stressor (r = .55, p \ .001). No other
moderation effects were not driven by either of these related relations between depression or social anxiety and affect at baseline
constructs. To do this, zero-order correlations were used to examine or stressor were found (all p’s [ .10). Where significant relations
the relation between self-reported depression (Beck Depression were found, depression and/or social anxiety were controlled for in
Inventory, Second Edition; Beck et al. 1996) and social anxiety moderation analyses. This did not impact any of our original results.
(Social Interaction Anxiety Scale; Mattick and Clarke 1998) symp- Rumination still did not moderate the NAB condition’s effect on
toms and all outcome variables. Depression was significantly subsequent negative affect (b = .03, t [84] = .19, p = .85 and
associated with post-training FC for positive stimuli (r = .24, b = .05, t [44] = .27, p = .78 when controlling for depression and
p = .046), but no other pre- or post-training eye tracking indices social anxiety, respectively), nor did it impact the PAB condition’s
(all p’s [ .10). Social anxiety was not related to any of the eye effect on FC for positive stimuli at post-training (b = .11,
tracking variables at pre- or post-training (all p’s [ .10). In addition, t [56] = .54, p = .59).

123
Cogn Ther Res (2014) 38:189–199 197

indices of attention biases or subjective affect in response an emotional reactivity effect, even in the absence of an
to the stressor. Though contrary to our hypotheses, such attention bias effect, that participants with low levels of
null findings are consistent with some past research, par- rumination assigned to the NAB condition demonstrated.
ticularly that which has employed CBM-A with dysphoric However, it is recommended that these results be replicated
stimuli (e.g., Baert et al. 2010). Still, as predicted, habitual before drawing firm conclusions about the impact of the
use of rumination was found to moderate training condition NAB training on subsequent emotional responding.
effects. Specifically, across all participants who underwent Despite evidence to support our moderation hypotheses,
the PAB versus the CTL training, rumination was related to it is acknowledged that the CBM-A procedures used in the
greater IO biases for positive stimuli at post-training. study did not produce effects that were robust enough to
Further, among participants who reported high levels of elicit change in attention or emotional responding across all
habitual rumination, those assigned to the PAB condition participants. There may be a couple of explanations for this
were more likely to demonstrate IO biases for positive lack of findings. First, it may be that biased attention
images at post-training than were high ruminators assigned functions differently in healthy versus clinical populations.
to the CTL condtion. Overall, individuals high in trait-level For example, it is suggested that patterns of engagement
rumination who were assigned to the PAB condition and disengagement with emotional stimuli are differen-
appeared to receive the most benefit from training proce- tially implicated in dysfunction across clinical and non-
dures. Rumination was not found to moderate effects on clinical populations (Ellenbogen et al. 2010). In addition, it
late-stage attention biases, nor did it moderate effects of the may be more difficult to train biases within healthy indi-
NAB training on post-training attention. viduals, perhaps because valenced environmental stimuli
Evidence for moderation effects was also found when (particularly negative/dysphoric stimuli) are less salient.
looking at participants’ subjective positive affect immedi- This hypothesis may, in part, explain why participants RTs
ately following the speech preparation task. When partic- did not change over the course of training trials. If this is
ipants endorsed high levels of rumination, affect did not the case, inducing attention biases in non-clinical popula-
differ as a function of study condition. Yet, among par- tions may require multiple training sessions.
ticipants reporting low levels of rumination, those who That the current study utilized a CBM-A paradigm
were assigned to either of the training conditions reported comprised of a single, 160-trial, training session should be
significantly more positive affect following the stressor considered a potential limitation of the study design. The
than participants assigned to the control condition. amount of training that is needed to produce effects on
Taken together, results indicate that CBM-A may not emotional reactivity remains unknown. While other
work the same for all individuals. Indeed, the extent to which research has effectively altered attention using single-ses-
training tasks can effectively alter attention and emotional sion paradigms of identical length (Amir et al. 2008),
responding appears to be dependent on the ways in which a others have relied on a greater number of trials (MacLeod
given individual typically responds to their emotions. et al. 2002) or multiple training sessions (Wells and Bee-
Interestingly, training effects were not confined to individ- vers 2010) to produce effects. Moreover, when a single
uals reporting either high or low levels of rumination. Rather, session of training has produced effects, it has not typically
increased use of rumination predicted greater biasing included a pre-post assessment of attention. Thus, it
towards positive stimuli, but decreased use of rumination remains unclear how long single-session training effects
predicted more adaptive stress responding. Though it may be expected to last. While the addition of the pre-post
remains unclear why this pattern of findings emerged, one assessment of attention should be noted as a methodolog-
possible explanation is that persons that ruminate continue to ical strength of the study, conducting the assessments at
process previously presented emotional stimuli at the separate time points before and after multiple sessions of
expense of regulating their current affective state (Yoon and training could increase the magnitude of observed training
Joormann 2012). Whereas this explanation fits well for the effects. In addition, though stimuli used during the assess-
PAB training, it does not explain the greater degree of ment of attention differed from that used during the training
positive affect experienced by participants in the NAB, as procedures, future studies may wish to employ different
compared to the CTL, condition. Though this greater degree stimuli at the pre- and post-training assessments to ensure
of positive affect among NAB participants was unexpected, that familiarity with emotional content does not affect
some previous research has found that training attention results. Lastly, while the CTL condition was designed to
towards threat is associated with attenuated anxiety, as serve as a comparison for both the NAB and PAB condi-
compared to a control condition (Klumpp and Amir 2010). It tions, an unintended consequence of the design was that the
is suggested that the mechanism of change in such cases is CTL differed from both conditions in the amount of expo-
not altered attention biases, but rather increased attentional sure it provided to either positive or negative stimuli. Thus,
control following training. This may explain the presense of the possibility that effects were due to differences in

123
198 Cogn Ther Res (2014) 38:189–199

exposure to valenced stimuli, rather than to altered attention Beck, A. T., Rush, A. J., Shaw, B. F., & Emery, G. (1979). Cognitive
biases cannot be ruled out. However, given that conditions therapy of depression. New York: Guilford Press.
Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the beck
did not differ in reported levels of positive or negative affect depression inventory-II. San Antonio, TX: Psychological
following training, this hypothesis seems unlikely. Corporation.
Of final note, regression analyses may have lacked Caseras, X., Garner, M., Bradley, B. P., & Mogg, K. (2007). Biases in
sufficient power to detect change. While the overall sample visual orienting to negative and positive scenes in dysphoria: An
eye movement study. Journal of Abnormal Psychology, 116,
was relatively large, there was a prominent loss of data in 491–497. doi:10.1037/0021-843X.116.3.491.
the collection of continuous focal attention via eye track- Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied
ing. Because of this, eye tracking analyses were conducted multiple regression/correlation analysis for the behavioral
on a subsample that was significantly smaller than the science (3rd ed.). Mahwah, NJ: Lawrence Erlbaum Associates.
Cronbach, L. J., & Furby, L. (1970). How we should measure
sample as a whole. Given this, drawing conclusions about ‘change’: Or should we? Psychological Bulletin, 74, 68–80.
null eye tracking effects is cautioned against. That rumi- doi:10.1037/h0029382.
nation was found to moderate early-stage attention biases De Lissnyder, E., Koster, E. H. W., Derakshan, N., & de Raedt, R.
for positive stimuli, however, speaks to the strength of this (2010). The association between depressive symptoms and execu-
tive control impairments in response to emotional and non-
relation. emotional information. Cognition and Emotion, 24, 264–280.
Despite its limitations, the current study adds to the doi:10.1080/02699930903378354.
extant literature by providing empirical support for rumi- Eckenrode, J., Ganzel, B., Henderson, C. R, Jr, Smith, E., Olds, D. L.,
nation as a moderator of CBM-A effects. This finding has Powers, J., et al. (2000). Preventing child abuse and neglect with
a program of nurse home visitation: The limiting effects of
obvious clinical implications, particularly for major domestic violence. Journal of the American Medical Associa-
depression, which is often associated with high levels of tion, 284, 1385–1391. doi:10.1001/jama.284.11.1385.
this construct. Future research should look to extend these Ellenbogen, M. A., Carson, R. J., & Pishva, R. (2010). Automatic
findings to CBM-A research with clinical samples. Doing emotional information processing and the cortisol response to
acute psychosocial stress. Cognitive, Affective, and Behavioral
so may allow us to better understand the link between Neuroscience, 10, 71–82. doi:10.3758/CABN.10.1.71.
rumination and attentional deployment for dysphoric Ellenbogen, M. A., Schwartzman, A. E., Stewart, J., & Walker, C.
environmental stimuli, as well as to identify a potential (2002). Stress and selective attention: The interplay of mood,
barrier in the implementation of CBM-A to reduce symp- cortisol levels, and emotional information processing. Psycho-
physiology, 39, 723–732. doi:10.1017/S0048577202010739.
toms in depressed populations. Gross, J. J., & Thompson, R. A. (2007). Emotion regulation:
Conceptual foundations. In J. J. Gross (Ed.), Handbook of
emotion regulation (pp. 3–26). New York: The Guilford Press.
Hallion, L. S., & Ruscio, A. M. (2011). A meta-analysis of the effect
References of cognitive bias modification on anxiety and depression.
Psychological Bulletin, 137, 940–958. doi:10.1037/a0024355.
Amir, N., Weber, G., Beard, C., Bomyea, J., & Taylor, C. T. (2008). Johnson, D. R. (2009). Goal-directed attentional deployment to
The effect of a single-session attention modification program on emotional faces and individual differences in emotion regulation.
response to a public-speaking challenge in socially anxious Journal of Research in Personality, 43, 8–13. doi:10.1016/j.jrp.
individuals. Journal of Abnormal Psychology, 117, 860–868. 2008.09.006.
doi:10.1037/a0013445. Joormann, J. (2010). Inhibition and emotion regulation in depression.
Armstrong, T., & Olatunji, B. O. (2012). Eye tracking of attention in Current Directions in Psychological Science, 19, 161–166.
the affective disorders: A meta-analytic review and synthesis. doi:10.1177/0963721410370293.
Clinical Psychology Review, 32, 704–723. doi:10.1016/j.cpr. Joormann, J., Dkane, M., & Gotlib, I. H. (2006). Adaptive and
2012.09.004. maladaptive components of rumination? Diagnostic specificity
Armstrong, T., Olatunji, B. O., Sarawgi, S., & Simmons, C. (2010). and relation to depressive biases. Behavior Therapy, 37,
Orienting and maintenance of gaze in contamination fear: Biases 269–280. doi:10.1016/j.beth.2006.01.002.
for disgust and fear cues. Behaviour Research and Therapy, 48, Klumpp, H., & Amir, N. (2010). Preliminary study of attention
402–408. doi:10.1016/j.brat.2010.01.002. training to threat and neutral faces on anxious reactivity to a
Baert, S., de Raedt, R., Schacht, R., & Koster, E. H. W. (2010). social stressor in social anxiety. Cognitive Therapy and
Attentional bias training in depression: Therapeutic effects Research, 34, 263–271. doi:10.1007/s10608-009-9251-0.
depend on depression severity. Journal of Behavioral Therapy Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (2005). International
& Experimental Psychiatry, 41, 265–274. doi:10.1016/j.jbtep. affective picture system (IAPS): Instruction manual and
2010.02.004. affective ratings. Technical Report A-6. University of Florida,
Beard, C., Sawyer, A. T., & Hofmann, S. G. (2012). Efficacy of Gainsville, FL.
attention bias modification using threat and appetitive stimuli: A MacLeod, C., & Mathews, A. (2012). Cognitive bias modification
meta-analytic review. Behavior Therapy, 43, 724–740. doi:10. approaches to anxiety. Annual Review of Clinical Psychology, 8,
1016/j.beth.2012.01.002. 189–217. doi:10.1146/annurev-clinpsy-032511-143052.
Beck, A. T., & Clark, D. A. (1997). An information processing MacLeod, C., Rutherford, E., Campbell, L., Ebsworthy, G., & Holker,
model of anxiety: Automatic and strategic processes. Behav- L. (2002). Selective attention and emotional vulnerability:
iour Research and Therapy, 35, 49–58. doi:10.1016/S0005- Assessing the causal basis of their association through the
7967(96)00069-1. experimental manipulation of attentional biases. Journal of

123
Cogn Ther Res (2014) 38:189–199 199

Abnormal Psychology, 111, 107–123. doi:10.1037/0021-843X. Segerstrom, S. C. (2001). Optimism and attentional bias for negative
111.1.107. and positive stimuli. Personality and Social Psychology Bulletin,
Mattick, R. P., & Clarke, J. C. (1998). Development and validation of 27, 1134–1343. doi:10.1177/01461672012710009.
measures of social phobia scrutiny fear and social interaction Strauss, G. P., & Allen, D. N. (2006). The experience of positive
anxiety. Behaviour Research and Therapy, 36, 455–470. doi:10. emotion is associated with the automatic processing of positive
1016/S0005-7967(97)10031-6. emotional words. The Journal of Positive Psychology, 1,
NIMH Center for the Study of Emotion and Attention. (1999). The 150–159. doi:10.1080/17439760600566016.
international affective picture system (IAPS). Gainsville, FL: The Treynor, W., Gonzalez, R., & Nolen-Hoeksema, S. (2003). Rumina-
Center for Research in Psychophysiology, University of Florida. tion reconsidered: A psychometric analysis. Cognitive Therapy
Preacher, K. J., Curran, P. J., & Bauer, D. J. (2004). Simple intercepts, and Research, 27, 247–259. doi:10.1023/A:1023910315561.
simple slopes, and regions of significance in MLR 2-way Wadlinger, H. A., & Isaacowitz, D. M. (2011). Fixing our focus:
interactions. Retrieved from http://www.quantpsy.org/interact/ Training attention to regulate emotion. Personality and Social
lca2_instructions.pdf. Psychology Review, 15, 75–102. doi:10.1177/1088868310365565.
Romens, S. E., & Pollack, S. D. (2012). Emotion regulation predicts Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and
attention bias in maltreated children at-risk for depression. validation of brief measures of positive and negative affect: The
Journal of Child Psychology and Psychiatry, 53, 120–127. PANAS scales. Journal of Personality and Social Psychology,
doi:10.1111/j.1469.2011.02474.x. 54, 1063–1070. doi:10.1037/0022-3514.54.6.1063.
Rueda, M. R., Posner, M. I., & Rothbart, M. K. (2005). The Wells, T. T., & Beevers, C. G. (2010). Biased attention and
development of executive attention: Contributions to the emer- dysphoria: Manipulating selective attention reduces subsequent
gence of self-regulation. Developmental Neuropsychology, 28, depressive symptoms. Cognition and Emotion, 24, 719–728.
575–594. doi:10.1207/s15326942dn2802_2. doi:10.1080/02699930802652388.
Sanchez, A., Vazquez, C., Marker, C., LeMoult, J., & Joormann, J. Yoon, L. K., & Joormann, J. (2012). Is timing everything? Sequential
(2012). Attentional disengagement predicts stress recovery in effects of rumination and distraction on interpersonal problem
depression: An eye-tracking study. Manuscript submitted for solving. Cognitive Therapy and Research, 36, 165–172. doi:10.
publication. 1007/s10608-010-9330-2.

123

You might also like