Professional Documents
Culture Documents
To cite this article: Tony T. Wells & Christopher G. Beevers (2010): Biased attention and dysphoria:
Manipulating selective attention reduces subsequent depressive symptoms, Cognition & Emotion, 24:4,
719-728
This article may be used for research, teaching, and private study purposes. Any substantial
or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the
contents will be complete or accurate or up to date. The accuracy of any instructions, formulae,
and drug doses should be independently verified with primary sources. The publisher shall not
be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or
howsoever caused arising directly or indirectly in connection with or arising out of the use of this
material.
COGNITION AND EMOTION
2010, 24 (4), 719728
BRIEF REPORT
Selective attention for dysphoric stimuli has been observed in individuals with depression and those
at risk for depression. To date, no studies have investigated the effects of directly manipulating
selective attention for dysphoric stimuli on depressive symptoms. Mild to moderately depressed
college students (N34) were randomly assigned to complete 4 sessions of either attention training
(AT) or no training (NT) during a two-week period. Participants completed self-reported
assessments of depressive symptoms at baseline, post-training, and follow-up. Participants in the
AT condition had a significantly greater decrease in depressive symptoms from baseline to follow-up
than participants in the NT condition. This group difference was mediated by change in attention
bias. Our findings suggest that biased attention may have a causal role in the maintenance of
depressive symptoms.
Correspondence should be addressed to: Tony T. Wells, Department of Psychology, University of Texas at Austin, 1 University
Station A8000, Austin, TX 787120187, USA. E-mail: wells@mail.utexas.edu
# 2009 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business 719
http://www.psypress.com/cogemotion DOI:10.1080/02699930802652388
WELLS AND BEEVERS
role in the maintenance of the disorder? attend to negative words compared with neutral
Earlier research indicated that attentional biases words reacted with more negative emotion to a
faded with remission of depressive symptoms stressful anagram task than participants who had
(e.g., Gotlib & Cane, 1987; McCabe & Gotlib, been trained to attend to neutral words. This was
1993), suggesting that selective attention for the first demonstration that selective attention
dysphoric stimuli might be an epiphenomenon could play a causal role in emotional vulnerability.
of depression. However, more recent studies have Importantly, this study also demonstrated that
observed biased processing of emotional informa- selective attention can be modified through atten-
tion outside the context of a mood episode. For tion training.
example, Beevers and Carver (2003) found that In the current investigation we tested the
increased attentional bias following a dysphoric hypothesis that biased attention plays a maintain-
mood induction, combined with life stress, pre- ing role in depression. More specifically, we
dicted increases in future dysphoria in non- investigated whether modifying selective attention
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
depressed college students. Joormann, Talbot, for dysphoric stimuli could reduce depressive
and Gotlib (2007) found a bias for sad faces symptoms. We used a dot-probe task to modify
following a sad mood induction in never de- selective attention for dysphoric stimuli in people
pressed girls at risk for developing depression. with elevated depressive symptoms. We also
Joormann and Gotlib (2007) found this same bias created a no training condition that was highly
for sad faces in individuals who had remitted from similar to the training condition with the excep-
major depression. tion that selective attention was assessed, but not
These studies suggest that selective attention is modified.
unlikely to simply be a by-product or symptom of We predicted that attentional bias for dysphoric
the disorder; however, these studies are correla- stimuli would significantly decrease in the training
tional and cannot rule out third-variable explana- condition and that no change would occur in the
tions. A more direct method for testing the causal no-training condition. We also predicted that
relationship between selective attention and main- participants assigned to the training condition
tenance of depression would be to examine the would experience a decrease in depressive symp-
effects of modifying attention on depressive toms but the participants in the no-training
symptoms. Papageorgiou and Wells (2000) used condition would not. Finally, we hypothesised
a case series design to test an attention-training that the association between attention training
procedure with clinically depressed individuals. and reductions in depressive symptoms would be
Using a therapist-guided procedure to train mediated by change in attentional bias.
attentional control, they found that attention
training improved participants’ depressive symp-
toms. However, the authors noted a number of METHOD
limitations which included small sample size, lack
Participants
of objective methods of assessment, and lack of a
control group. While this provides preliminary Participants were 34 undergraduate students with
evidence, no randomised, controlled study has mild to moderate symptoms of depression who
examined whether modifying biased attention completed the study as part of a research require-
reduces depression symptoms. ment for an introduction to psychology course.
MacLeod, Rutherford, Campbell, Ebsworthy, Participants completed the short form of the Beck
and Holker (2002) were able to affect non-clinical Depression Inventory (BDI; Beck & Steer, 1993)
participants’ emotional vulnerability with a com- during mass pre-testing. Participants whose scores
puter-administered dot-probe task that was were above 7 on the short form of the BDI were
designed to modify selective attention to negative contacted about participating in the current study.
stimuli. Participants trained to preferentially Of the 345 who were contacted, 50 attended the
baseline laboratory session. Among those who (described below) during the 2 weeks following
were scheduled to participate, 16 were no longer the baseline laboratory session. They were also
experiencing mild-to-moderate symptoms of scheduled for a post-training assessment session 2
depression (i.e., a BDI-II score between 9 and weeks from the baseline screening appointment
25) and were dismissed from the study. In order and a follow-up assessment session that occurred
to select a sample with at least mild symptoms of 4 weeks after the baseline session. Participants
depression, we used a BDI-II cut-off score of 9, were awarded experimental credit for their intro-
which has been shown to maximise sensitivity in ductory psychology course for participation at
selecting for depressed mood in a college sample baseline, training sessions, and post-training.
(Sprinkle et al., 2002). Of the remaining 34 who This provided their full allotment of credit, so
met study criteria, 91% completed all 4 training participants were offered $5 and entry in a raffle
sessions and the post-training session assessment to win $75 for participation in the follow-up. At
and 53% completed the follow-up session. Full the baseline, post-training, and follow-up sessions
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
and partial data obtained from all 34 participants (each separated by two weeks), participants com-
were used in analyses. pleted the BDI-II. Participants also completed
the Beck Anxiety Inventory (BAI; Beck, Epstein,
Materials Brown, & Steer, 1988) at baseline.
A set of 12 pairs of faces, each from a different Participants were told that the purpose of the
actor (6 males and 6 females) and each expressing study was to examine the relationship between
sad and neutral emotions, was selected from the mood and attention. It is important to note that
Pictures of Facial Affect (POFA) photo set the term ‘‘training’’ was not mentioned to the
developed by Ekman and Friesen (1976). We participants. They were told that they would be
also selected 20 images from the International returning over several sessions to help validate an
Affective Picture System (IAPS; Lang, Bradley, attention task. They were not given any informa-
& Cuthbert, 2005), 10 displaying a neutral scene tion that they might be undergoing attention
or image and 10 displaying a dysphoric scene or training or that it could improve their mood until
image.1 An example of a dysphoric IAPS scene they were debriefed at the end of the experiment.
includes several children with torn and tattered This ensured that the participants were unaware
clothes facing the viewer. The children are of the study purpose as well as experimental
crowded together and appear to be in great condition. In addition, participants were assigned
distress with several children clearly crying. An randomly to the attention training or no training
example of a neutral scene includes a middle-aged conditions. The experimenters were blind to the
man sitting outside reading a newspaper. Faces conditions of participants until completion of
and images were used rather than word stimuli the experiment. Before debriefing at the end of
due to the success of other recent studies in the study, participants were asked to guess the
detecting a bias using these types of stimuli (e.g., purpose of the task. No participants correctly
Eizenman et al., 2003; Joormann & Gotlib, guessed the purpose or hypothesis of the study.
2007). The images were presented on a 20-inch
LCD PC monitor. Design
Dot-probe task. The set of 22 image pairs (12
Procedure
POFA face pairs and 10 IAPS image pairs) was
Upon meeting the inclusion criteria, participants presented in 9 blocks in each session for a total of
were scheduled to return for 4 training sessions 196 trials per session. Each POFA pair consisted of
1
We used the following IAPS images. Neutral: 2102, 2393, 2745, 2850, 5390, 5731, 7009, 7041, 7053, and 7493. Dysphoric:
2141, 2205, 2276, 2455, 2700, 2703, 2799, 2900, 9421, and 9530.
one sad face and one neutral face of the same actor. two asterisks. Participants completed 11 practice
One IAPS neutral image was always paired with trials (5 IAPS and 6 POFA pairs) using neutral
one IAPS dysphoric image in the IAPS pairs; neutral pairs. These practice trials used an equal
however, the pairings were randomly chosen dur- probability distribution for the probe for all
ing each new block. Total time to complete each participants. Participants repeated the practice
training session was approximately 25 minutes. until they responded accurately to at least 9 of
Each trial consisted of a white fixation cross on the 11 practice trials.
a black background in the centre of the screen for
1500 ms, followed by an image pair. The size of Training manipulation. At the first training
each POFA image was approximately 12.7 session, participants were randomly assigned by
19 cm (subtending approximately 128188 of computer to attention training (AT) or no train-
the visual angle) and each IAPS image was ing (NT) conditions. The conditions differed only
approximately 17.112.7 cm (approximately in the probability of the probe location following
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
168128). The pictures in each pair were the neutraldysphoric stimulus pair. In the NT
approximately 21 cm apart when measured from condition, the probe appeared in the location of
each centre and were presented in the left and the neutral stimulus and the dysphoric stimulus
right halves of the screen. This gave a visual angle with equal probability (50%). In the AT condi-
of approximately 208 between the centres of each tion, however, the probe appeared in the location
image and 108 between the image centre and the of the neutral stimulus 85% of the time. We used
fixation cross. 85% rather than 100% for two reasons: to allow
POFA pairs were presented for 3000 ms and for assessment of attention bias during the task
IAPS pairs were presented for 4500 ms. The and to keep the intent of the study from being
relatively longer stimulus durations compared to transparent. Sixty standard dot-probe trials (i.e.,
standard dot-probe administrations (e.g., Gotlib equal probability of the probe appearing in the
et al., 2004) were intended to allow participants location of each stimulus) in the first and last
time to more fully process the content of the training sessions were used to calculate attention
stimuli, as longer durations may allow for more bias in both the AT and NT conditions.
elaborated processing and greater activation of This 85% probe distribution in the AT condi-
relevant schemata (Mogg & Bradley, 2005). tion served as the critical manipulation whereby
Following the offset of the images, a small single the participants were putatively trained to allocate
or double white asterisk probe on a black back- their attention preferentially to the neutral
ground appeared in the location of one of the stimuli. This was intended to train participants
images and remained on the screen until the to allocate their attention away from the dysphoric
participant pressed a corresponding response-box stimuli.
button to indicate whether they saw one or two
asterisks. The computer recorded the latency and
accuracy of each response. Each type of emotional RESULTS
stimulus (sad or neutral) and probe stimulus
(single or double asterisk) appeared on each side Mean age for the sample was 19.1 (SD1.7).
of the screen with equal probability. There were no significant differences between
training conditions in the proportion of women
Task presentation. Participants were told that (Ns: AT13, NT11), x2(1, N24)B1, ns,
their goal was to determine whether they saw number of dropouts, x2(1, N16)B1, ns, age,
one or two asterisks as quickly and accurately as t(32)B1, ns, BAI score, t(32)B1, ns, or initial
possible. They used their index finger to press a BDI-II score, t(32)1.12, p.27. Mean BDI-II
button marked * when they saw one asterisk and scores (and standard deviations) for the sample as
used their middle finger to press ** when they saw a whole, the AT condition, and the NT condition
were 16.1 (4.8), 17.1 (5.5), and 15.2 (4.1), for each participant, were considered outliers and
respectively. Mean BAI scores were 11.6 (6.7), were eliminated. This resulted in a loss of 1.64%
11.9 (7.5), and 11.3 (6.1), respectively. Two of the data and did not differ between conditions,
participants in the NT condition were taking t(31)1.35, p.19.
antidepressant medication during the study; and
no participants in the AT condition reported Bias scores. In line with the procedure of Gotlib
taking antidepressant medication. Analyses with- et al. (2004), attentional bias scores were calcu-
out or with these two participants from the NT lated for each participant for each of the four
condition resulted in an identical pattern of training sessions using the following equation:
results. Therefore, they were included in all
analyses reported below. Attentional bias score
Of the 34 participants who were randomised to 1
[(RpLeRpRe)(LpLeLpLe)] 1
a training condition, 31 (Ns: AT14, NT17) 2
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
Figure 1. Bias score for dysphoric stimuli at sessions 1 and 4 completed within a two-week time period for the attention training (AT) and
no training (NT) conditions. Error bars represent the standard error of the mean.
19 NT AT
17
BDI
15
13
11
9
Baseline Post-training Follow-up
Time
Figure 2. Self-reported depression (BDI-II) measured at baseline, post-training, and follow-up for the attention training (AT) and no
training (NT) conditions. Error bars represent the standard error of the mean. Figure includes estimated data.
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
score at session 1 and baseline BDI-II to their bias score at session 4 and replaced it with bias
respective subsequent assessments. score at session 1. This model did not achieve
This initial model achieved an adequate fit, good fit, x2(2)10.84, pB.01, x2/df5.42,
x2(3)5.57, p.13, x2/df1.85, CFI.92, CFI.28, SRMR.19. Taken together, our
SRMR.10. More importantly, assignment to results suggest that change in attentional bias
the AT condition (training away from dysphoric mediates the association between attention train-
images) was strongly associated with reductions in ing and depressive symptoms at follow-up.
biased attention for dysphoric stimuli at session 4,
controlling for bias at session 1. Further, a lower
bias score at session 4 was associated with fewer DISCUSSION
depression symptoms at follow-up, controlling for
baseline depressive symptoms. This model ex- This study examined whether an attention-training
plained 53.1% of the variance in depression at task designed to reduce a bias for dysphoric stimuli,
follow-up. compared to a similar task that had no training
To evaluate the specificity of our model, we component, would lead to mood improvement in
tested an alternative mediational model. Specifi- participants with mild-to-moderate levels of de-
cally, we examined whether bias score at session 1 pressive symptoms. Consistent with our predic-
mediated the relationship between attention tions, attention training (AT) led to a reduction in
training and BDI-II at follow-up. We thus tested depressive symptoms while participants in the no
the same model as in Figure 3, but we removed training condition (NT) experienced no significant
−.09 .57*
Figure 3. Results from path analyses indicating that attentional bias at session 4 (controlling for attention bias at session 1) mediates the
association between attention training condition (coded 0 NT, 1AT) and depressive symptoms at follow-up (controlling for baseline
depressive symptoms).
change. Effect sizes for attention training were in bolster the argument that, rather than being a
the medium-to-large range using definitions pro- by-product of depression, selective attention has a
posed by Cohen (1988). The interaction between causal role in the maintenance of the disorder.
time and training group explained approximately The results of this study should be interpreted
one third of the variance in BDI-II score. The with some limitations in mind. Perhaps the most
effect size for change in depressive symptoms in the significant limitation is the dropout rate before
AT group was medium for baseline to post- the follow-up session; 53% of participants com-
training and large for baseline to follow-up. In pleted the follow-up session. However, we took
contrast, the effect size for change in depressive several steps to ensure reliability of the reported
symptoms in the no-training condition was close to effects. Namely, our method of estimating miss-
zero. Finally, attention training also reduced biased ing data (FIML) is considered a ‘‘state-of-the-art’’
attention for dysphoric stimuli, which in turn approach (Schafer & Graham, 2002). FIML has
mediated the association between attention train- been shown to be superior to removing cases with
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
ing and reductions in depressive symptoms. missing data (i.e., listwise deletion) for a several
To our knowledge, this is the first study to statistical techniques (Enders, 2001; Enders &
examine the effects of attention training on Bandalos, 2001). In addition, Little and Rubin
depressive symptoms using a no-training control (2002) and Schafer (1997) have both suggested
condition and carefully controlling for expectancy that FIML data estimation methods can be used
effects. The current study controlled for expec- reliably when up to 90% of the data are missing
tancy effects in that participants and experimen- for a particular variable; however, in such circum-
ters were unaware of whether participants were stances, analyses involving that variable rely on
assigned to the active or inert training conditions. model and missing data assumptions to a greater
In fact, our study went a step further to reduce degree than when missing data are minimal. It
expectancy effects by not revealing until the end of should also be noted that our analyses using
the study that there were two separate training listwise deletion and last observation carried
conditions and that one condition could alter forward produced very similar results to those
depressive symptoms. This enabled us to effec- using FIML estimated data, indicating consis-
tively rule out expectancy effects as a factor in the tency between results produced by different
improvement observed in the AT condition. methods of data estimation as well as non-
Importantly, we observed a decrease in atten- estimated data. Future studies with larger sample
tional bias for dysphoric stimuli in the AT group sizes and better participant retention are needed
but not the NT group. Participants displayed an to replicate our findings.
increased bias towards neutral stimuli and away A second limitation is that we did not have an
from dysphoric stimuli over time in the AT ‘‘assessment only’’ control group. An alternative
condition, but relatively little change in the NT interpretation for our results is that the AT group
condition. Thus, change in attentional bias improved naturally while some component of the
appeared to mediate the effect of training condi- NT condition prevented those participants from
tion on change in depression symptoms. improving. In other words, though the NT
Our study illuminates the relationship between condition involved a standard dot-probe task, it
depression and attentional bias for dysphoric may not have been completely inert. Though this
stimuli. The current study is the first that we design limitation is common to studies of atten-
are aware of to find that experimentally manip- tion training (e.g., MacLeod et al., 2002; Li, Tan,
ulating attentional bias for dysphoric stimuli can Qian, & Liu, 2008), future studies should imple-
lead to symptom reduction. The results of this ment an assessment-only control to address this.
study suggest that selective attention for dysphoric A third limitation is that the same stimuli were
stimuli plays an important role in the maintenance used in the attention training and the assessment
of depressive symptoms. These findings also of attention bias at the first and fourth training
sessions. Though this is not an optimal procedure, Beck, A. T., & Steer, R. A. (1993). Beck Depression
MacLeod et al. (2002) found that their attention Inventory: Manual. San Antonio, TX: The Psycho-
training produced no significantly different post- logical Corporation.
training effects on new stimuli compared to Beevers, C. G., & Carver, C. S. (2003). Attentional
bias and mood persistence as prospective predictors
stimuli on which the participant had been trained.
of dysphoria. Cognitive Therapy and Research, 27,
This indicates that attention training seems to
619637.
affect both previously viewed and new stimuli Cohen, J. (1988). Statistical power analysis for the
similarly. However, future studies should use behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence
different stimuli for attention training to reliably Erlbaum Associates, Inc.
assess the external validity of change in bias Eizenman, M., Yu, L. H., Grupp, L., Eizenman, E.,
scores. The current study did not assess major Ellenbogen, M., Gemar, M., et al. (2003). A
depression and so some participants would not naturalistic visual scanning approach to assess selec-
have met criteria for clinical depression. Addi- tive attention in major depressive disorder.
Downloaded by [Duke University Libraries] at 10:01 13 March 2013
Li, S., Tan, J., Qian, M., & Liu, X. (2008). Continual Mogg, K., & Bradley, B. P. (2005). Attentional bias in
training of attentional bias in social anxiety. generalized anxiety disorder versus depressive
Behaviour Research and Therapy, 46, 905912. disorder. Cognitive Therapy and Research, 29, 2945.
Little, R. J. A. (1988). A test of missing completely at Papageorgiou, C., & Wells, A. (2000). Treatment of
random for multivariate data with missing values. recurrent major depression with attention training.
Journal of the American Statistical Association, 83(4), Cognitive and Behavioral Practice, 7, 407413.
11981202. Schafer, J. L. (1997). Analysis of incomplete multivariate
Little, R. J. A., & Rubin, D. B. (2002). Statistical data. Boca Raton, FL: Chapman & Hall/CRC.
analysis with missing data (2nd ed.). Hoboken, NJ: Schafer, J. L., & Graham, J. W. (2002). Missing data:
Wiley. Our view of the state of the art. Psychological
MacLeod, C., Rutherford, E., Campbell, L., Ebs- Methods, 7(2), 147177.
worthy, G., & Holker, L. (2002). Selective attention Sprinkle, S. D., Lurie, D., Insko, S. L., Atkinson, G.,
Jones, G. L., Logan, A. R., et al. (2002). Criterion
and emotional vulnerability: Assessing the causal
validity, severity cut scores, and testretest reliability
basis of their association through the experimental
Downloaded by [Duke University Libraries] at 10:01 13 March 2013