Content uploaded by Emily Holmes
Author content
All content in this area was uploaded by Emily Holmes on Jun 28, 2016
Content may be subject to copyright.
Developing Interpretation Bias Modification as a “Cognitive Vaccine” for
Depressed Mood: Imagining Positive Events Makes You Feel Better Than
Thinking About Them Verbally
Emily A. Holmes, Tamara J. Lang, and Dhruvi M. Shah
University of Oxford
Two interpretation bias modification experiments found that mental imagery vs. verbal processing of
positive material have differential emotional effects. In Experiment 1, participants were instructed to
imagine positively resolved auditory descriptions or to listen to the same events while thinking about
their verbal meaning. Increases in positive mood and bias were greater in the imagery than in the verbal
condition, replicating E. A. Holmes, A. Mathews, T. Dalgleish, and B. Mackintosh (2006). An emotional
vulnerability test showed that imagery (relative to the verbal condition) protected against a later negative
mood induction. Experiment 2 created 2 new verbal conditions aimed to increase or reduce verbal
comparisons. Results suggest making unfavorable comparisons with the highly positive material might
be partially responsible for the inferiority of the verbal condition in Experiment 1. The findings
demonstrate that imagery can play a key role in cognitive bias modification procedures and thus that task
instructions are crucial. Imagining a positive event can make you feel better than thinking about the same
event verbally. The authors propose that recruiting imagery will be useful in therapeutic innovations to
develop a “cognitive vaccine” for depressed mood.
Keywords: mental imagery, depression, positive affect, interpretation, cognitive bias modification
Depressed and anxious people suffer from underlying biases in
their style of thinking as if “filtering” information in a negative
way. For example, a negative interpretation bias refers to the
tendency to negatively interpret ambiguous stimuli (Butler &
Mathews, 1983). Recent computerized cognitive bias modification
(CBM) techniques that target interpretation (CBM-I) suggest pos-
sibilities for modifying such biases (Grey & Mathews, 2000;
Mathews & Mackintosh, 2000; Mathews & MacLeod, 2002).
CBM-I encourages individuals to repeatedly constrain potentially
ambiguous interpretations in a particular direction (positive or
negative) that can, over time, habitually bias the interpretation of
fresh ambiguous information. CBM-I has also been shown to
modify vulnerability to experiencing anxiety to a later stressor
(e.g., Mackintosh, Mathews, Yiend, Ridgeway, & Cook, 2006).
Using CBM-I to train negative versus benign biases allows us to
examine the etiology of bias in clinical disorders (Mathews &
MacLeod, 2005) and delineate influencing factors. Using CBM-I
to train in a benign or positive direction offers exciting possibilities
for treatment innovation. Although CBM-I methods aiming to
improve clinical biases are blossoming, the optimal ingredients are
yet to be identified.
Our first attempts at using “nonnegative” or benign CBM-I
material failed to produce improvements in anxious mood or bias
(Holmes & Mathews, 2005). Thus, Holmes, Mathews, Dalgleish,
and Mackintosh (2006) developed overtly positive training mate-
rial, resulting in the first successful test of positive CBM-I. This
training material comprised auditory descriptions that were am-
biguous initially but consistently resolved with positive outcomes.
Two instruction conditions were compared. Mental imagery com-
pared with verbal instructions had more powerful effects on (a)
emotion (increases in positive affect and decreases in anxiety) and
on (b) interpretation bias, even when administered after an interval
to remove any difference between conditions in state mood that
could otherwise be argued to drive effects. Within the imagery
condition, positive affect increased over training. Unexpectedly,
within the verbal condition pre- to post-, the CBM-I training phase
mean positive affect scores decreased significantly, and negative
bias increased. Thus, verbal instructions were not only inferior to
imagery in producing positive emotion, but within this condition,
mood deteriorated over training. Our objectives in the present
study were to replicate this effect and to explore links to depressed
bias and mood.
It could be argued that our findings in favor of imagery are at
odds with other CBM-I methods showing positive effects of “ver-
bal training.” However, we believe this is not the case. The original
CBM-I paradigm using verbal scripts by Mathews and Mackintosh
(2000) used imagery instructions—“imagine yourself in the situ-
ation.” Most subsequent studies have included imagery instruc-
tions and obtained mood effects (using auditorily presented verbal
scripts: Holmes & Mathews, 2005; Holmes et al., 2006; Mackin-
Emily A. Holmes, Tamara J. Lang, and Dhruvi M. Shah, Department of
Psychiatry, University of Oxford, Oxford, England.
This research was supported by a Royal Society Dorothy Hodgkin
Fellowship and in part by Economic and Social Research Council Grant
RES-061-23-0030 and John Fell OUP Grant PRAC/JF awarded to Emily
A. Holmes. We thank Andrew Mathews for formative discussion about the
ideas in this article and Anna Coughtrey for her help in preparing this
manuscript for publication.
Correspondence concerning this article should be addressed to Emily
Holmes, Department of Psychiatry, University of Oxford, Warneford Hospital,
Oxford OX3 7JX, United Kingdom. E-mail: emily.holmes@psych.ox.ac.uk
Journal of Abnormal Psychology © 2009 American Psychological Association
2009, Vol. 118, No. 1, 76– 88 0021-843X/09/$12.00 DOI: 10.1037/a0012590
76
tosh et al., 2006; using visually presented verbal scripts: Mathews,
Ridgeway, Cook, & Yiend, 2007; Murphy, Hirsch, Mathews,
Smith, & Clark, 2007; Yiend, Mackintosh, & Mathews, 2005). It
is of interest that Salemink, van den Hout and Kindt (2007) did not
mention imagery in their method and found bias differences with
only a marginal emotional effect.
The seminal article by Mathews and Mackintosh (2000) dem-
onstrated that active generation of meaning was critical to produc-
ing emotional changes. Mathews and MacLeod (2002) argued that
generation probably has this effect via imagery (or access to
personal memories in similar form). This was confirmed by
Holmes and Mathews (2005). In the only other related study we
are aware of with no imagery instructions, Hirsch, Mathews, and
Clark (2007) again did not find an anxiety change over training.
However, training-congruent effects were obtained when partici-
pants later imagined ambiguous items and on anticipated anxiety
to imagined social stress. There is rich evidence for changing
semantic bias without explicit imagery instructions or mood
change, as in the use of homographs in CBM-I (rather than verbal
scripts). Here, spontaneous imagery during training seems implau-
sible. For example, Wilson, MacLeod, Mathews, and Rutherford
(2006) obtained effects both on bias and a challenge task (stressful
films). We assume that verbal-semantic CBM-I can prime va-
lenced semantic meanings but has little emotional impact unless
imagery is also involved either during training or subsequently
under circumstances when the priming influences some type of
image content.
Our research has also used the CBM paradigm for a broader line
of inquiry about whether mental imagery has a special relationship
with emotion. Holmes et al. (2006) suggested that mental imagery
can have greater effects on positive emotion than verbal processing
of the same material. Why might this be the case? Clinically
powerful negative mental imagery is a feature of several psycho-
logical disorders (Hackmann & Holmes, 2004), but it can also be,
at least in part, positive as in substance cravings (Kavanagh,
Andrade, & May, 2005) or suicidal ideation (Holmes, Crane,
Fennell, & Williams, 2007). As we have argued (Holmes &
Mathews, 2005), unlike verbal information, imagery has percep-
tual correspondence to direct sensory experience, “as if” it were
really happening (Kosslyn, Ganis, & Thompson, 2001). Imagery
may directly provoke emotion as a real percept might, both for
positive and negative emotion. By mimicking real-life perceptions,
imagery can also aid access to representations of related emotional
autobiographical memories (Conway, 2001) with their accompa-
nying emotional tone. Similar machinery is activated when imag-
ining the future as when remembering the past (Schacter, Addis, &
Buckner, 2007). Thus, as for sensory perceptual autobiographical
memories, prospective imagery may also accrue emotional effects.
Imagined events are more likely to be confused with reality than
verbal descriptions (Hyman & Pentland, 1996). In contrast, posi-
tive verbal information may be less believable and more readily
contrasted with other (disconfirmatory) information accessible in
rich verbal semantic networks. This may be highly adaptive in
other domains (such as debating a point of information) but proves
unhelpful for being cheered by positive information.
Why might verbal processing of positive material not only be
inferior to imagery but also possibly lead to paradoxical effects? In
psychopathology, depression and anxiety are associated with ver-
bal processing in the form of rumination and worry. Clinical
theories across disorders converge on the idea that verbal/abstract
processing may in the short term reduce negative affect, but it may
also be maladaptive. The interacting cognitive subsystems theory
(Teasdale & Barnard, 1993) addresses depressive rumination. The
reduced concreteness theory (Sto¨ber & Borkovec, 2002) focuses
on worry, suggesting that such abstract verbal thinking is associ-
ated with reduced imagery and thus used to avoid distressing
imagery. Relatedly, Hayes and Gifford (1997) suggested the cost
of using verbal language to avoid negative affect can later be a
paradoxical increase in negative emotion. There is evidence that
inducing rumination and worry in nonclinical participants de-
creases positive affect (McLaughlin, Borkovec, & Sibrava, 2007).
The maladaptive subcomponent of rumination—brooding —is
defined as a “passive comparison of one’s current situation with
some unachieved standard” (Treynor, Gonzalez, & Nolen-
Hoeksema, 2003, p. 256). A potential account of the verbal con-
dition findings of Holmes et al. (2006) is that participants were
verbally comparing their current situation with “unachieved stan-
dards” highlighted by the overtly positive training material. In-
deed, thinking about discrepancies between how one actually is
(actual self) with either how one would ideally like to be (ideal
self) or feel one should be (ought self) has been found to relate to
depression and anxiety, respectively (Strauman & Higgins, 1987).
One can of course also make comparisons using imagery, but
this may be less automatic and more effortful, requiring active
switching between images (with deliberate imagery generation
known to take seconds; Cocude, Charlot, & Denis, 1997), or
switching between imagery and verbal modes. Furthermore, im-
agery can be highly absorbing; for example, flashbacks to trauma
are difficult to dismiss from the mind (Brewin & Holmes, 2003).
In the positive domain, individuals can indulge in positive imag-
inal daydreams, reveries, and fantasies (Kavanagh et al., 2005)
without contradicting them until they come to an end (and then
switch to comparative or verbal processing). Perhaps these absorb-
ing and believable properties of imagery (at least in the moment)
enhance the ability to benefit from even unrealistic and highly
positive information.
Although our initial positive CBM-I research began in the
context of anxiety, negative interpretation bias has held a central
role in traditional cognitive-behavioral theories of depression
(Beck, 1976). Depressed and dysphoric participants display more
negative interpretation bias than controls, as indicated in numerous
studies reviewed by Bisson and Sears (2007). Their study also
questioned the robustness of this effect, arguing that some previous
findings may be accounted for by demand. However, their review
omitted two relevant studies in which methods that are less sus-
ceptible to these concerns—namely, the scrambled sentences test
under a cognitive load, are used (Rude, Valdez, Odom, & Ebra-
himi, 2003) and an eye blink response to ambiguous words (Law-
son, MacLeod, & Hammond, 2002). Moreover, Rude et al. (2003)
found bias predicted future depression. On balance, the negative
interpretation bias associated with depressed mood appears to be
an important therapeutic target.
Other research suggests that there may be problems related to
mental imagery in depression, with high levels of negative intru-
sive imagery (Kuyken & Brewin, 1994), a lack of positive imagery
(Williams et al., 1996), and a preponderance of verbal thought
when ruminating (Fresco, Frankel, Mennin, Turk, & Heimberg,
2002). People with depression have an overgeneral memory bias
77
SPECIAL SECTION: POSITIVE BIAS TRAINING
(Williams et al., 2007), which is antithetical to generating con-
crete, specific mental imagery. On the basis of converging litera-
ture demonstrating negative interpretation bias and lack of positive
mental imagery in depression, we predicted that our imagery-based
positive CBM-I paradigm would be beneficial for alleviating de-
pressed mood.
Experiment 1
We sought to (a) replicate the results of Holmes et al. (2006) in
a larger sample and test whether positive imagery CBM-I would
(b) extend to depressive bias and (c) transfer to a depression-
relevant test of emotional vulnerability (negative mood induction).
As our interpretation bias measure using ambiguous test descrip-
tions (Holmes & Mathews, 2005; Holmes et al., 2006) was devel-
oped in the context of anxiety, it may not tap into a depression-
relevant negative bias. Rude and colleagues (2003) found a
relationship between bias on the Scrambled Sentences Test (SST;
Wenzlaff, 1993) and subsequent depressive symptoms. Under cog-
nitive load, the proportion of negative solutions on the SST pre-
dicted diagnoses of depression 18 –24 months later. We therefore
tested whether our overtly positive CBM-I might influence later
performance on the SST.
If CBM is to be effective, then we need to demonstrate transfer
of effects on later tests of emotional vulnerability. For anxiety,
relevant “stressor/challenge” tests include an insoluble anagram
task (MacLeod, Rutherford, Campbell, Ebsworthy, & Holker,
2002; Salemink et al., 2007) and stressful film clips (Wilson et al.,
2006). In the context of depressed mood, a relevant test of later
emotional vulnerability is a negative mood induction.
Differential activation theory (Lau, Segal, & Williams, 2004)
was developed to account for the finding that some people remain
vulnerable to a reoccurrence of depressive symptoms, even when
typical cognitive vulnerability markers (e.g., dysfunctional atti-
tudes) indicate recovery. This theory asserts that it is not the
“resting level” of such attitudes that is critical after recovery, but
how easily these attitudes are reactivated by small changes in
mood. This rationale has been extended to suicidality in depression
(Williams, Barnhofer, Crane, & Beck, 2005). Because even small
mood changes are thought to reactivate potentially harmful think-
ing styles, we argue it may be useful to develop tasks that protect
against mood deterioration. We therefore tested whether our train-
ing conditions differentially affect mood following a negative
mood induction.
Our key hypotheses were as follows:
Hypothesis 1: We predicted that positive CBM-I would be
more effective in the imagery condition compared with the
verbal condition, with relatively greater reductions in anxiety,
complementary increases in positive affect, and more positive
interpretation bias, replicating Holmes et al. (2006). We pre-
dicted this difference would extend to the SST, a depression-
relevant measure of interpretation bias.
Hypothesis 2: We sought to replicate the unexpected finding
from Holmes et al. (2006) of mood deterioration in the verbal
condition over the training phase (specifically, increase in
state anxiety, reduction in positive affect, and increased neg-
ative interpretation bias).
Hypothesis 3: Conversely, within the imagery condition
alone, we predicted significant improvements in mood and
bias.
Hypothesis 4: We predicted that compared with participants
in the verbal condition, those in the imagery condition would
be “protected” from mood deterioration over a later negative
mood induction.
Method
Overview
Using the positive CBM-I paradigm described by Holmes et al.
(2006), 100 auditory scenarios were presented. The scenarios were
initially ambiguous as to their outcome then yielded consistently
positive resolutions. Participants were instructed to either imagine
the events or listen to them while thinking about their verbal
meaning. Participants were randomized to either imagery or verbal
conditions. The same training materials were used in both condi-
tions. Mood measures (Positive and Negative Affect Schedule
[PANAS]; Watson, Clark, & Tellegen, 1988; and the Spielberger
State-Trait Anxiety Inventory [STAI]; Spielberger, Gorsuch,
Lushene, Vagg, & Jacobs, 1983) were completed pretraining,
immediately posttraining, and after a 10-min filler task. To assess
interpretation bias, emotional valence ratings for ambiguous test
scenarios were completed pretraining and after the filler task, and
the SST was completed after the filler task. A negative mood
induction was delivered, followed by a final administration of
mood measures and manipulation check questions.
Participants
The 40 participants were recruited through advertisements in the
local town and paid a small fee for participation; see Table 1 for
characteristics.
Materials
Positive training paragraphs. One hundred descriptions used
by Holmes et al. (2006) were implemented. Each one described a
situation that had a positive emotional outcome. The descriptions
were read aloud in the same female voice (each lasting approxi-
Table 1
Characteristics of Participants in Experiment 1
Characteristic
Imagery
(n⫽20)
Verbal
(n⫽20)
MSDM SD
Age (years) 29.05 8.68 32.90 13.94
Gender (%)
Women 50 60
Men 50 40
STAI Trait 33.65 8.54 34.80 7.92
BDI-II 4.00 3.04 4.75 3.73
SUIS 3.75 0.63 3.66 1.12
Note. STAI ⫽State-Trait Anxiety Inventory; BDI-II ⫽Beck Depression
Inventory-II; SUIS ⫽Spontaneous Use of Imagery Scale.
78 HOLMES, LANG, AND SHAH
mately 10 –13 s) and digitally recorded using Cool Edit 2000
software (Syntrillium Software Corporation; Phoenix, AZ). The
descriptions were presented stereophonically via headphones using
E-Prime software (Version 1.1.4.1; Psychology Software Tools
Inc.; Pittsburgh, PA). An example is as follows: “You have started
an evening class which is tough going. You are determined to
succeed, and after a while, it becomes much easier and more
enjoyable”(resolution in italics). The initial part of the scenario
was designed to be ambiguous in the sense that it could also imply
a negative outcome (finding class difficult). A second example was
“You are starting a new job that you very much want. You think
about what it will be like and feel extremely optimistic” (resolution
in italics). The initial part of the scenario, despite being generally
positive, was still intended to be somewhat ambiguous, as it could
be resolved with overtly positive outcomes such as “feel pleased”
or even “feel concerned.” There was a 2-s gap after each descrip-
tion. Items were randomized throughout five training blocks, each
consisting of 20 paragraphs.
Consistent with Holmes et al. (2006), in order to focus partici-
pants on their assigned task, after each training paragraph (and 2-s
gap), they either rated vividness of imagery (“How vividly could
you imagine the situation that was described?”) or their ability to
comprehend the description (“How difficult was it to understand
the meaning of the description?”), depending on condition (imag-
ery vs. verbal). Ratings were made on a 7-point scale ranging from
1(not at all)to7(very). Task instruction reminders were given
between blocks, with short breaks allowed between these blocks.
Filler task. During the 10-min interval after training, partici-
pants performed an unrelated filler task to eliminate mood differ-
ences between groups prior to the bias tests. A series of classical
music extracts was played randomly, each lasting 40 s. Participants
rated how pleasant they found each extract on a scale ranging from
1(extremely unpleasant)to9(extremely pleasant).
Measures of interpretation bias. The SST (Wenzlaff, 1993)
instructed participants to unscramble a list of 20 scrambled sen-
tences in 2.5 min under a cognitive load. Participants were re-
quired to order five of the six words to create a grammatically
correct sentence by placing a number from 1 to 5 over them. This
constrained participants to select a positive or negative sentence.
For example “good feel very bad I usually” could be unscrambled
as “I usually feel very bad” (negative) or “I usually feel very good”
(positive). Prior to unscrambling, a six-digit number was presented
for 5 s and then hidden for 10 s. Participants were instructed to
hold this number in mind during the task and then write it down.
The aim of this concurrent task was to provide a cognitive load.
Participants were given three benign practice sentences and in-
structed to complete the task as quickly as possible. After the task,
97.2% of participants reported the number accurately. The SST
was included as an index of state bias following the experimental
manipulation, whereas it is more typically used as a traitlike
measure.
The ambiguous test descriptions comprised 10 positively resolv-
able paragraphs used by Holmes et al. (2006). The descriptions
were ambiguous because potential emotional outcomes were im-
plied but not explicitly described. They were presented before and
after the training phase, without specific instructions as to encod-
ing. Participants rated the emotional valence of each description
using a 9-point scale ranging from 1 (extremely unpleasant)to9
(extremely pleasant). For example, “You buy a new outfit for a
party. Other people’s reactions show how you look.”
Questionnaire measures. The two versions of the STAI (Spiel-
berger et al., 1983) index trait and state anxiety separately. These
widely used measures have satisfactory reliability and validity
(Spielberger et al., 1983).
State positive affect was measured using the Positive Affect
subscales of the PANAS (Watson et al., 1988). Negative affect
items were excluded. These comprise 21 items—the basic positive
emotion scales (Joviality, Self-Assurance, Attentiveness) as well
as the Serenity subscale of Watson and Clark (1994). Items were
administered with short-term time instructions that anchored re-
sponses to feelings within the past few minutes.
The tendency to use imagery in everyday life was measured
using the Spontaneous Use of Imagery Scale (SUIS; Reisberg,
Pearson, & Kosslyn, 2003). This 12-item questionnaire is rated on
a 5-point scale ranging from 1 (never appropriate)to5(always
completely appropriate) (e.g., “When I think about visiting a
relative, I almost always have a clear mental picture of him or
her”.) Reisberg et al. (2003) found a positive relationship between
the Vividness of Visual Imagery Questionnaire (Marks, 1973) and
the SUIS, suggesting they measure a related concept.
The Beck Depression Inventory-II (BDI; Beck, Steer, & Brown,
1996) measured baseline depressive symptoms. This widely used
measure of depression has robust reliability and validity (Beck et
al., 1996).
Negative mood induction. Participants read negative Velten
(1968) statements presented on Microsoft PowerPoint (2003)
while listening through headphones to depressing music—the or-
chestral introduction by Prokofiev, Russia under the Mongolian
Yoke by Prokofiev from the film Alexander Nevsky, played at
half-speed as used in Williams et al. (2005). There were 30
statements repeated for 2.5 min rather than for 8 min to allow for
a greater variation in response. Two visual analogue scales (VASs)
were used (Williams et al., 2005) to measure how “sad” or
“happy” participants were feeling “at this moment” by marking a
10-cm line ranging from not at all to extremely.
Manipulation check ratings. Participants made three ratings of
their experience listening to the sentences during the training phase
to assess (a) imagery processing—“How much did you find your-
self thinking in images, i.e., in mental pictures and sensory im-
pressions?”; (b) verbal processing—“How much did you find
yourself verbally analyzing the meaning of the sentences?”, and (c)
concentration difficulties—“How much of the time did you find it
difficult to focus on your task, i.e. your attention wandered and you
found it difficult to concentrate?” Responses were rated on a
9-point scale ranging from 1 (not at all)to9(all the time).
Procedure
After giving their informed consent to the experiment, partici-
pants were randomly assigned to either the imagery or the verbal
condition. They completed the BDI-II, SUIS, PANAS, and STAI
(Time 1). They listened to the first administration of the ambigu-
ous test descriptions, presented in random order, and rated each for
emotional valence. The experimenter then read instructions for the
assigned condition.
The instructions were identical to those used by Holmes et al.
(2006). In the imagery condition, participants were given a brief
79
SPECIAL SECTION: POSITIVE BIAS TRAINING
practice task in which they were asked to imagine cutting a lemon.
This task is routinely used in our studies to clarify what is meant
by “using mental imagery” and create an experience of deliberate
image generation. Participants were given four sample descriptions
and asked to “imagine each event as happening to themselves” as
the description unfolds while describing their mental image out
loud, with the final example administered via computer. The
experimenter explained that maintaining a focus on their images
would help in answering the questions that followed.
In the verbal condition, participants completed an equivalent of
the “cutting the lemon” task, with instructions to focus on the
meaning of each description as they heard it. Participants were
given four sample descriptions, with instructions to “concentrate
on the words and meaning as the description unfolds.” They were
told not to imagine the situation. A final example was administered
using the computer. The experimenter stated that focusing on the
words and meaning of each description would help them to answer
the questions that followed.
Participants were then given 100 auditory training descriptions
in 5 randomized blocks of 20. For any participant, depending on
their assigned condition, all descriptions were followed by a 2-s
pause, after which a vividness rating (imagery condition) or com-
prehension rating (verbal condition) was completed. STAI and
PANAS were repeated at the end of the training phase (Time 2).
An interval of 10 min was given and filled by a neutral task. Mood
measures were administered at the end of the interval to check
mood equivalence between groups (Time 3; STAI, PANAS, and
VAS). Participants then completed a second administration of the
ambiguous test descriptions and the SST. The negative mood
induction was then administered, followed by mood assessments
using the PANAS and VAS (Time 4). Participants then completed
the manipulation checks ratings. Finally, they were debriefed and
thanked for their participation.
Results
All data was analyzed via SPSS Version 13.0.
Comparison of Participants in Imagery and Verbal
Conditions
The groups were comparable in terms of gender,
2
(1, N⫽
40) ⫽0.10, p⫽.75, age; trait STAI, BDI-II, SUIS, and Time 1
state STAI, PANAS, and ambiguous test description ratings (ts⬍
1.2, ps⬎.25); see Tables 1 and 2.
Mood Change From Pretraining to Immediately
Posttraining
State anxiety. We predicted that participants in the imagery
condition would show greater reductions in state anxiety than
those in the verbal condition. We tested this hypothesis by using
a mixed model analysis of variance (ANOVA), with a grouping
variable of condition (imagery vs. verbal) and the within-
subjects variable of time (Time 1: pretraining vs. Time 2:
posttraining). There was no main effect of time or condition
(Fs⬍1). However, as predicted, there was a significant inter-
action of time and condition, F(1, 38) ⫽15.20, p⬍.001,
p
2
⫽
.29. This interaction was decomposed by separate paired sam-
ples ttests of change over time within conditions. There was a
significant decrease in anxiety in the imagery condition, t(19) ⫽
2.28, p⫽.04, d⫽1.31 (mean change ⫽⫺2.85, SD ⫽5.60).
There was a significant increase in anxiety after verbal training,
t(19) ⫽3.22, p⫽.005, d⫽0.67 (mean change ⫽⫹4.25, SD ⫽
5.91).
Positive affect. We tested the hypothesis that participants in
the imagery condition would show greater increases in positive
affect than those in the verbal condition by using a similar
ANOVA to that described above. There was no main effect of time
(F⬍1) or condition, F(1, 38) ⫽1.02, p⫽.32. As predicted, there
was a significant interaction of time and condition, F(1, 38) ⫽
10.67, p⫽.002,
p
2
⫽.22. The imagery group showed a significant
increase in positive affect, t(19) ⫽5.80, p⬍.001, d⫽0.37 (mean
change ⫽⫹4.95, SD ⫽3.82). In contrast, the decrease in positive
affect (mean change ⫽⫺2.90, SD ⫽10.05) in the verbal condition
did not reach significance, t(19) ⫽1.29, p⫽.21, d⫽0.18.
State Anxiety and Positive Affect After the Filler Task
An independent ttest confirmed there were no significant dif-
ferences between groups in positive affect or state anxiety (ts⬍1,
ps⬎.38) at Time 3 (see Table 2). Thus, there were no longer
significant differences in mood between groups prior to adminis-
tration of interpretation bias tests.
Table 2
Means and Standard Deviations for State Mood Measures, Bias
Measures, and Manipulation Checks per Condition
Measure
Imagery
(n⫽20)
Verbal
(n⫽20)
MSDMSD
Mood measures
State STAI, Time 1 29.85 8.49 27.15 5.52
State STAI, Time 2 27.00 5.94 31.40 7.12
State STAI, Time 3 27.10 7.63 28.40 6.04
PANAS, Time 1 73.10 13.25 72.40 14.45
PANAS, Time 2 78.05 12.94 69.50 18.62
PANAS, Time 3 78.10 16.08 73.60 17.22
PANAS, Time 4 73.20 17.96 62.70 20.08
VAS sad, Time 3 5.65 8.79 2.00 2.51
VAS sad, Time 4 10.05 14.82 17.95 21.44
VAS happy, Time 3 75.20 17.61 73.85 17.03
VAS happy, Time 4 70.20 18.94 56.45 28.77
Bias measures
Ambiguous test
descriptions, Time 1
6.23 0.92 6.21 0.91
Ambiguous test
descriptions, Time 3
6.50 1.17 6.23 1.02
Scrambled
Sentences Test,
Time 3
0.064 0.085 0.144 0.118
Manipulation checks
Use of imagery 8.85 0.49 4.40 1.47
Use of verbal 4.20 3.14 6.20 2.63
Note. Time 1 ⫽pretraining; Time 2 ⫽immediately posttraining; Time
3⫽after 10-min filler task posttraining; Time 4 ⫽after mood induction;
STAI ⫽State-Trait Anxiety Inventory; PANAS ⫽total positive affect
score from the Positive and Negative Affect Schedule; VAS ⫽Visual
Analogue Scale ratings.
80 HOLMES, LANG, AND SHAH
Interpretation Bias
The SST. A total negativity score was produced by calculating
the number of negatively valenced sentences over the total number
of sentences completed (Rude et al., 2003). Two participants in the
imagery condition scored more than 3 standard deviations from the
mean and were removed from the analysis (Tabachnick & Fidell,
1996). Using an independent samples ttest, as predicted, negativ-
ity scores were greater in the verbal than in the imagery condition,
t(36) ⫽2.37, p⫽.02, d⫽0.77 (see Table 2). Additional analyses
indicated that this effect was no longer significant when outliers
were included (t⬍1, p⬎.33).
Ambiguous test descriptions. We predicted that, compared
with participants who were given imagery training, those given
verbal training would rate scenarios more negatively after training.
Using a similar ANOVA to those described above, there was no
significant main effect of time or condition (Fs⬍1). The inter-
action between time and training condition was not significant
(F⬍1).
Response to Negative Mood Induction
We predicted that compared with those in the verbal condition,
participants in the imagery condition would show less deteriora-
tion in mood following the negative mood induction. We tested
this hypothesis by using a mixed model ANOVA, with a grouping
variable of condition (imagery vs. verbal) and the within-subjects
variable of time (Time 3: premood induction vs. Time 4: postmood
induction).
The PANAS. There was a main effect of time, F(1, 38) ⫽
42.74, p⬍.001,
p
2
⫽.53, qualified by a significant interaction of
time and condition, F(1, 38) ⫽6.16, p⫽.02,
p
2
⫽.14. There was
no main effect of condition, F(1, 38) ⫽1.84, p⫽.18,
p
2
⫽.05. We
tested our directional hypothesis by comparing the reduction in
PANAS scores from Time 3 to Time 4 between groups using an
independent samples ttest. There were greater reductions in the
verbal group (mean change ⫽⫺10.90, SD ⫽8.33) than in the
imagery group (mean change ⫽⫺4.90, SD ⫽6.88), t(38) ⫽2.48,
p⫽.02, d⫽0.79.
VAS sadness. There was a main effect of time, F(1, 38) ⫽
13.94, p⫽.001,
p
2
⫽.27, and no main effect of condition (F⬍1).
The interaction of time and condition, F(1, 38) ⫽4.49, p⫽.004,
p
2
⫽.11, was significant. The increase in sadness in the verbal
group (mean change ⫽15.95, SD ⫽21.48) was significantly
greater than the increase in the imagery group (mean change ⫽
4.40, SD ⫽11.53), t(38) ⫽2.12, p⫽.04, d⫽0.67.
VAS Happiness. There was a main effect of time, F(1, 38) ⫽
16.10, p⬍.001,
p
2
⫽.30; a significant interaction of time and
condition, F(1, 38) ⫽4.94, p⫽.03,
p
2
⫽.12; and no main effect
of condition, F(1, 38) ⫽1.55, p⫽.22,
p
2
⫽.04. There were
greater decreases in happiness in the verbal group (mean change ⫽
⫺17.40, SD ⫽22.09) than in the imagery group (mean change ⫽
⫺5.00, SD ⫽11.63), t(38) ⫽2.22, p⫽.03, d⫽0.70.
The negative mood induction results may be partially accounted
for by a differential emotional response to the ambiguous stimuli
presented within this same interval. Thus, to check that mood
change was not a result of bias change, we repeated the ANOVA
for the PANAS, but with the SST score as a covariate. The
interaction of time and condition remained significant, F(1, 37) ⫽
5.57, p⫽.02,
p
2
⫽.13, when SST was covaried. This is consistent
with the suggestion that the response to the negative mood induc-
tion was not accounted for by completing the SST. A similar
pattern of results was obtained for both VAS scales.
Manipulation Checks
Participants in the verbal condition reported using more verbal
processing during the training phase, t(38) ⫽2.19, p⫽.04, d⫽
0.69, and those in the imagery condition reported using more
imagery, t(38) ⫽12.88, p⬍.001, d⫽4.06 (see Table 2). There
was no significant difference between groups for reported diffi-
culty concentrating on the task (t⬍1, p⬎.6; M⫽2.83, SD ⫽
2.19).
Discussion
Experiment 1 replicated key findings from Holmes et al. (2006)
of the benefits of imagery compared with verbal CBM-I, using a
larger sample. Significantly for the CBM-I paradigm, we addition-
ally demonstrated transfer to a novel test of emotional vulnerabil-
ity—a depressive mood induction. Within the imagery condition
alone, there were significant improvements in mood compared
with baseline (positive affect and state anxiety). In contrast, pos-
itive CBM-I with verbal instructions led to not just a lack of mood
improvement, but to an increase in anxiety over the training phase.
Within the verbal condition, we did not replicate a significant
increase in negative interpretation bias on the ambiguous test
scenarios, although cautions about the previous findings had been
raised by Holmes et al. (2006). However, depressive interpretation
bias (as gauged by the SST) was more positive after imagery than
verbal CBM-I, with means in a similar range to those reported by
Rude et al. (2003).
As predicted, relative to the verbal condition, the imagery con-
dition was “protected” from mood deterioration on the negative
mood induction. These results provide a stringent test of CBM-I on
later emotional vulnerability because the conditions contrasted
were all positive rather than varying in training valence, for ex-
ample, benign versus negative, as in Wilson et al. (2006). The
results also extend a test of emotional vulnerability to the novel
domain of depressed mood. It is possible that the negative mood
induction result may partially be accounted for by a differential
emotional response to the ambiguous stimuli presented within this
same interval (the bias tests). However, when covarying for SST,
the interaction was retained. Future research in which further
checks (e.g., post-SST mood ratings) are included could ensure
that differential emotional reactions to the ambiguous stimuli
themselves were not responsible for our results.
We sought to test other potential accounts of our results. The
relative mood deterioration associated with the verbal condition
may be due to differences between conditions in fatigue. However,
this is unlikely to give a full account given the lack of difference
in perceived concentration difficulties between conditions. Fur-
thermore, manipulation check ratings were consistent with adher-
ence to the appropriate condition instructions, although experi-
menter demand cannot be ruled out. Finally, there was no
significant difference in mood between groups prior to the bias
tests, which could have otherwise influenced results. The critical
reasons for the inferiority of the verbal condition remain to be
81
SPECIAL SECTION: POSITIVE BIAS TRAINING
tested. Nevertheless, it will be clinically advantageous to develop
CBM paradigms that do not lead to mood deterioration over the
course of a session, as this could reduce compliance.
In conclusion, Experiment 1 provides evidence that instructions
about how to focus on positive material in a CBM paradigm can
significantly alter impact on mood, despite exposure to identical
positive stimuli. We have replicated key results from Holmes et al.
(2006) and successfully extended findings to depressive bias
(SST). Crucially, differential CBM-I effects transferred to a de-
pressive mood induction, indicating that positive training using
imagery relative to verbal scripts was protective. Reading the
self-relevant negative statements (e.g., “The future seems just one
string of problems”) during the mood induction may be well
matched to our (self-relevant) training descriptions in terms of
transfer-appropriate processing (Hertel, 2002).
We propose that these early results suggest that imagery CBM-I
may hold promise for development as a cognitive vaccine against
later depressed mood. By cognitive vaccine we mean that analo-
gous to medical vaccines, the CBM paradigm may be harnessed to
boost peoples’ resistance to experiencing a later emotional chal-
lenge. Clinical translation will benefit from extension to further
behavioral tasks in the laboratory and to an examination of re-
sponses to stressful real-world events. Researchers also need to
refine what properties of CBM will best accrue resistance to later
stress. The present data suggests that using imagery is worthy of
further investigation.
Of interest, if mode of processing alone (imagery vs. verbal)
rather than acquired positive bias was transferred, then we would
not expect the present mood induction results. Our previous work
has shown that negative imagery training (vs. verbal) can increase
(rather than protect against) negative mood (Holmes & Mathews,
2005), suggesting that imagery effects are valence specific. This
pattern of results would not appear to be accounted for by frame-
works that focus only on mode of processing (e.g., abstract vs.
concrete; Watkins, 2004; Watkins & Moulds, 2005). Future ex-
planations in CBM will need to continue to examine the combi-
nation of mode of processing (e.g., imagery vs. verbal) with
valence of bias (positive vs. negative).
As we have argued (Holmes & Mathews, 2005; Holmes et al.,
2006), it is likely that using imagery is an active rather than
incidental component of the original procedure developed by
Mathews and Mackintosh (2000). That is, the original “imagine
yourself in the situation” instructions were critical. An important
implication for CBM is that in our paradigm, typical training
effects found on measures of affect and emotional bias were not
achieved in our paradigm with verbal instructions alone. This
finding has potential clinical implications as it suggests that at-
tending to positive information using a processing style engen-
dered by our verbal processing instructions may even decrease
rather than increase positive mood. This could have unfortunate
ramifications for the use of verbal-only discussion of positive
events in cognitive therapy.
To date, a limitation of our experiments is insufficient explora-
tion of the nature of imagery or verbal processing. There are
clearly different varieties of both imagery and verbal processing
styles. We need to identify the active ingredients responsible for our
findings. From a clinical perspective, it seems crucial to know how we
might modify verbal instructions so that when presented with positive
material, verbal processing might be less likely to worsen mood.
Experiment 2
What aspect of verbal processing caused the mood deterioration
during the positive CBM-I training phase seen in Experiment 1 and
Holmes et al. (2006)? Experimental debriefing indicated that ver-
bal condition participants in Experiment 1 reported thoughts like
“things never work out like this for me.” This alerted us to the
possibility that participants may have unfavorably compared the
outcome of the overtly positive scenarios with their own, not as
(extremely) positive experiences. Although we did not expect our
nonclinical sample to have chronically negative experiences, nei-
ther were they expected to have unusually positive lives. For
example, consider the scenario “It’s a rainy day and you go outside
with your umbrella. As the rain falls around you, you notice your
step quickens and you whistle and feel surprisingly cheerful.”
Comparing this outcome with what typically happens in real life,
participants may dislike or only tolerate bad weather rather than
enjoy it, so that the comparison would be negative.
We have seen that for clinical conditions, verbal processing can
be associated with negative emotion. Clearly not all verbal pro-
cessing will yield negative emotional effects. What cognitive
mechanisms might account for paradoxical effects of verbally
processing positive information? We suggest the most compelling
explanation is that a process of making verbal comparisons with
positive information known as “evaluation” can produce negative
emotional consequences (Markman & McMullen, 2003). Evalua-
tion is defined as “an evaluative mode of thinking characterized by
the use of information about the standard as a reference point
against which to evaluate one’s present standing” (Markman &
McMullen, 2003, p. 245). Evaluative processing is predicted to
produce affective contrast (i.e., if comparing with more positive
information, negative affect would result, and vice versa).
Support for this comes from the social psychology literature that
explains that social comparisons made between worse off (down-
ward) and better off (upward) others often produce positive and
negative emotions, respectively (Morse & Gergen, 1970). Further-
more, the counterfactual thinking literature argues that whereas
downward counterfactual thinking (considering worse alterna-
tives) improves affect, upward counterfactual thinking (consider-
ing better alternatives) worsens affect (Markman, Gavanski, Sher-
man, & McMullen, 1993).
These theories suggest ways in which being presented with
positive material might make participants more negative (i.e., our
verbal processing instructions may provoke “evaluative” compar-
isons). That is, participants may have been making comparisons
between themselves and the overtly positive scenarios that may be
causing the increase in negativity found in the verbal condition
(but be less likely in imagery) of Experiment 1 and Holmes et al.
(2006). To test this proposal, we created two new verbal conditions
aimed to either increase or decrease the amount of comparisons
being made. They were both based on the existing verbal condition
instructions of Experiment 1, plus either (a) additionally instruct-
ing participants to compare each scenario with how things are for
them in reality (the “verbal comparisons condition”) or (b) remov-
ing the reference to “focus on the meaning” in the instructions, as
well as reducing the time available to make comparisons (the
“verbal reduced-comparisons condition,” so-called to reflect the
intention of the manipulation). The original imagery condition of
Experiment 1 was included as a control.
82 HOLMES, LANG, AND SHAH
In the verbal reduced-comparisons condition, we reduced the
time available to make comparisons by removing the short gap
after each auditory training scenario and speeding the scenarios
(while preserving comprehension). The latter was motivated by
Pronin and Wegner’s (2006) study, in which participants read
statements at either double or half normal reading speed. Increased
speed was associated with greater increases in positive mood.
Our key hypotheses were as follows:
Hypothesis 1: Following positive CBM-I, the verbal compar-
isons condition, compared with either the imagery or the
verbal reduced-comparisons condition, would produce
greater increases in state anxiety (STAI) with complementary
effects on the PANAS and on bias.
Hypothesis 2: Within the verbal comparisons condition alone
there would be significant deterioration in mood and bias over
the training phase. Within the imagery condition, mood and
bias over the training phase would improve.
Method
Participants
A further 60 participants were recruited using the same proce-
dure as before, see Table 3.
Materials
Positive training paragraphs. The same 100 descriptions were
used as in Experiment 1. In the verbal reduced-comparisons con-
dition, to reduce time available for making comparisons, the sce-
narios were played 20% faster, and the 2-s gap after each descrip-
tion was removed. The sound files were speeded using WavePad
software (Version 3.05, NCH Swift Sound; Canberra, Australia).
The mean duration of each letter was approximately 59 ms (pre-
viously 74 ms). In addition, the word meaning was eliminated
from the instruction “focus on the words and meaning.”
After each paragraph, participants rated either vividness of
imagery (as in Experiment 1), ability to comprehend the scenario
(“How difficult was it to understand the description?”), or differ-
ence from reality (“How different was this description compared to
how things really are for you in reality?”). Ratings were made on
5-point scales ranging from 1 (not at all vivid/difficult/different)to
5(extremely vivid/difficult/different).
Filler task. This was the same as Experiment 1.
Ambiguous test descriptions (interpretation bias). This was
the same as Experiment 1.
Questionnaire measures. The BDI-II, SUIS, PANAS, and
STAI were administered at baseline as before. Additionally, dif-
ferences between groups in general tendency were tested to make
comparisons using the Self-Guide Strength measure (Higgins,
Shah, & Friedman, 1997). Participants listed six characteristics
representing their ideal and ought self. They rated how much they
would like (extent ratings) and how much they felt they actually
possessed (actual ratings) these attributes. One ideal discrepancy
score and one ought discrepancy score was calculated by subtract-
ing extent ratings from actual ratings.
Manipulation check ratings. Imagery, verbal use, and concen-
tration during training were assessed as in Experiment 1. In addi-
tion, participants were asked “How much did you find yourself
comparing the scenarios with how things are for you are in reality
as you were listening to the sentences?” using the same rating
scale.
Procedure
The procedure was similar to Experiment 1, with the following
exceptions:
1. The Self-Guide Strength measure was administered at
baseline.
2. There were two verbal conditions with instructions mod-
ified from Experiment 1. Specifically, “Do your best to
focus on the words and meanings while listening to the
sentences and compare the information with how things
really are for you in reality” (verbal comparisons condi-
tion) and “Do your best to focus on the words while
listening to the sentences” (verbal reduced-comparisons
condition).
Table 3
Characteristics of Participants in Experiment 2 per Condition
Characteristic
Imagery
(n⫽20)
Verbal reduced-
comparisons
(n⫽20)
Verbal comparisons
(n⫽20)
MSDMSDMSD
Age (years) 24.40 7.23 25.15 8.16 25.30 7.07
Gender (%)
Women 60 75 65
Men 40 25 35
STAI Trait 38.50 9.25 37.30 11.82 37.30 10.20
BDI-II 7.20 4.95 8.50 7.56 7.25 7.18
SUIS 4.00 0.61 3.73 0.87 3.98 0.76
Ideal discrepancy 4.05 1.61 4.35 2.60 4.90 2.27
Ought discrepancy 3.45 1.96 3.90 1.89 4.05 1.57
Note. STAI ⫽State-Trait Anxiety Inventory; BDI-II ⫽Beck Depression Inventory-II; SUIS ⫽Spontaneous
Use of Imagery Scale; the ideal and ought discrepancy ratings were taken from the Self-Guide Strength measure.
83
SPECIAL SECTION: POSITIVE BIAS TRAINING
3. In the verbal reduced-comparisons condition, the audi-
tory training stimuli were played 20% faster, and the gap
after each description was removed.
4. The SST and negative mood induction were excluded as
our focus in Experiment 2 was on the impact of verbal
comparisons on previous training effects of mood and
bias (rather than on transfer effects to depression related
measures per se).
Results
Comparison of Participants in Verbal Comparisons,
Verbal Reduced-Comparisons, and Imagery Conditions
There were no significant differences between conditions in
gender,
2
(2, N⫽60) ⫽1.05, p⫽.59; see Table 3. Conditions
were comparable in terms of age, trait anxiety (STAI), imagery
(SUIS), positive affect (PANAS), depression (BDI-II), and ambig-
uous test scenarios (Fs ⬍1); see Table 4. There were no differ-
ences on a measure potentially related to our experimental manip-
ulation, the Self-Guide Strength measure (F⬍1); see Table 3.
Mood Change From Pretraining to Immediately
Posttraining
State anxiety. We predicted that participants in the verbal
comparisons condition would experience greater increases in anx-
iety relative to either the verbal reduced-comparisons or imagery
condition. Repeating the analysis of state anxiety used in Experi-
ment 1, we found no main effects (Fs⬍1). As predicted, there was
a significant interaction between time and condition, F(2, 57) ⫽
6.31, p⫽.002,
p
2
⫽.18. Using independent samples ttests, as
expected, anxiety increased significantly more in the verbal com-
parisons than in the verbal reduced-comparisons group (mean
change ⫽⫹3.8, SD ⫽5.85 vs. ⫺1.55, SD ⫽7.49), t(38) ⫽2.52,
p⫽.016, d⫽0.80. The predicted difference between the verbal
comparisons and imagery group was also significant (mean change
in imagery ⫺2.7, SD ⫽4.91), t(38) ⫽3.81, p⬍.001, d⫽1.20.
There was no significant difference between the imagery and
verbal reduced-comparisons groups (t⬍1, p⬎.56).
Paired samples ttests revealed, as predicted, a significant in-
crease in anxiety over training within the verbal comparisons
condition, t(19) ⫽2.91, p⫽.009, d⫽0.49, and a significant
decrease in anxiety within the imagery condition, t(19) ⫽2.46, p⫽
.024, d⫽0.43. There was no significant change in anxiety within
the verbal reduced-comparisons condition, (t⬍1, p⬎.36).
Notably, the inclusion of a Bonferroni correction (a conservative
protection against an inflated alpha caused by computing three
between-condition comparisons) in our analyses (.05 / 3 ⫽.017)
did not change the pattern of results.
Positive affect. We hypothesized that the verbal comparisons
condition compared with both verbal reduced-comparisons and
imagery conditions would produce greater decreases in positive
affect. Repeating the analysis for positive affect used in Experi-
ment 1, we found a main effect of time, F(1, 57) ⫽10.24, p⫽.002,
p
2
⫽.15, with positive affect decreasing, and no main effect of
condition, F(2, 57) ⫽1.36, p⫽.27. There was a significant
interaction between time and condition, F(2, 57) ⫽3.99, p⫽.03,
p
2
⫽.12.
There was a significant difference between the verbal compar-
isons and imagery groups (mean change ⫽⫺7.00, SD ⫽8.99 vs.
mean change ⫽⫹0.90, SD ⫽10.94), t(38) ⫽2.50, p⫽.017, d⫽
0.79. Unexpectedly, there was no significant difference between
the verbal reduced-comparisons (mean change ⫽⫺5.60, SD ⫽
8.18) and verbal comparisons groups (t⬍1, p⬎.6). The difference
between imagery and verbal reduced-comparisons groups was
significant, t(38) ⫽2.13, p⫽.04, d⫽0.67. Paired samples ttests
showed a significant decrease in positive affect within the verbal
comparisons, t(19) ⫽3.48, p⫽.002, d⫽0.53, and verbal reduced-
comparisons group, t(19) ⫽3.06, p⫽.006, d⫽0.36. Unexpect-
Table 4
Means and Standard Deviations for State Mood Measures and Bias Measures per Condition
Measure
Imagery
(n⫽20)
Verbal reduced-
comparisons
(n⫽20)
Verbal comparisons
(n⫽20)
MSDM SD MSD
Mood measures
State STAI, Time 1 33.95 8.11 35.60 12.16 31.40 7.06
State STAI, Time 2 31.25 8.16 34.05 9.60 35.20 8.45
State STAI, Time 3 30.95 6.72 32.40 9.77 32.50 6.76
PANAS, Time 1 69.30 16.14 65.40 15.16 71.00 12.79
PANAS, Time 2 70.20 15.18 59.80 15.64 64.00 13.54
PANAS, Time 3 67.75 15.35 59.10 15.81 66.45 15.23
Bias measures
Ambiguous test descriptions, Time 1 6.39 0.72 6.20 0.96 6.31 0.99
Ambiguous test descriptions, Time 3 6.50 0.91 6.01 1.02 6.21 0.89
Manipulation checks
Use of imagery 8.00 0.73 3.25 1.74 3.53 1.21
Use of verbal 2.50 1.36 7.45 1.05 7.45 1.04
Use of comparisons 4.90 2.43 3.75 2.10 8.25 0.79
Note. Time 1 ⫽pretraining; Time 2 ⫽immediately posttraining; Time 3 ⫽after 10-min filler task posttraining; STAI ⫽State-Trait Anxiety Inventory;
PANAS ⫽total positive affect score from the Positive and Negative Affect Schedule.
84 HOLMES, LANG, AND SHAH
edly, there was no significant increase in positive affect following
imagery training (t⬍1, p⬎.7).
If we include a Bonferroni correction, it affects only the (non-
critical) difference between the verbal reduced-comparisons and
imagery condition ( p⫽.04) rendering it nonsignificant.
State Anxiety and Positive Affect After the Filler Task
A one-way ANOVA confirmed that there were no significant
differences between groups at Time 3 in either state anxiety (F⬍
1) or positive affect, F(2, 57) ⫽1.82, p⫽.17.
Ambiguous Test Descriptions
Using a mixed model ANOVA, there were no main effects
(Fs⬍1). The interaction between time and training condition did
not reach significance, F(2, 57) ⫽1.67, p⫽.20.
Manipulation Checks
The four ratings were compared using one-way ANOVAs and
decomposed with independent samples ttests. For mean scores,
see Table 4. First, reported use of imagery was higher in the
imagery than in both verbal conditions ( ps⬍.001). There were no
differences between the two verbal conditions in reported use of
imagery (t⬍1, p⬎.56). Second, both verbal conditions reported
using more verbal processing than the imagery condition ( ps⬍
.001), with no differences between verbal groups (t⬍1, p⬎.99).
Third, reported use of verbal comparisons was higher in the verbal
comparisons than in the other conditions ( ps⬍.001). There were
no differences for use of verbal comparisons between the other two
conditions, t(38) ⫽1.60, p⫽.12.
Finally, there was no significant difference between the three
conditions for difficulty concentrating on the training task (F⬍1;
M⫽6.48, SD ⫽1.84). The pattern of results did not change when
a Bonferroni correction was applied.
Discussion
In Experiment 2, we sought to examine a potential mechanism
to account for the finding in the present Experiment 1 and Holmes
et al. (2006) that exposure to overtly positive material in a verbal
instruction condition was inferior to an imagery condition, and led
to mood worsening over the training phase. We proposed one
possibility was that participants were unfavorably comparing the
very positive training material with their personal (not so consis-
tently positive) experiences. Such comparisons with positive in-
formation were predicted to lead to mood deterioration (Markman
& McMullen, 2003; Strauman & Higgins, 1987). We therefore
created two new verbal conditions that aimed to either increase or
decrease comparisons. Our critical result was that positive training
in the verbal comparisons condition led to greater increases in
anxiety than in both other conditions. Thus, for state anxiety, the
verbal reduced-comparisons condition ameliorated the negative
emotional impact relative to imagery. This supports our prediction
that a feature of this new condition— comparative verbal process-
ing— contributed, at least in part, to previous findings.
The analogous pattern of results for positive affect seem less
clear, though because Watson et al. (1988) have demonstrated the
independence of positive and negative affect, this may be unsur-
prising. Unexpectedly, the positive mood increase within the im-
agery condition alone did not reach significance. Apart from the
changes detailed to create the two new verbal conditions, only one
further modification was made to Experiment 2—a measure of
self-discrepancy was added prior to the training phase. Unfortu-
nately, it may have primed participants to make comparisons by
explicitly requiring them to do so just before training began.
Indeed, our manipulation check ratings showed that the mean
comparison score in the imagery condition reached the midrange
(see Table 4). Future research needs to test our prediction that
effective imagery conditions are characterized by lower ratings of
comparisons.
There was no difference between conditions in difficulty con-
centrating during training. Manipulation check ratings were con-
sistent with adherence to the appropriate condition instructions,
though experimenter demand cannot be ruled out. For reasons
given earlier, the test of emotional vulnerability in Experiment 1
was omitted in Experiment 2, and future studies should include
this. Caution should still be taken in drawing conclusions, as it is
always possible some further factor incidental to our experimental
manipulation could account for the pattern of results. It would be
interesting to compare the verbal-reduced comparisons condition
with a new imagery condition matched for speed and length.
However, because deliberate imagery generation is slow, reducing
time in an imagery condition may be confounded with reduced
imageability during CBM-I (Cocude et al., 1997). Future research
should experimentally manipulate comparisons within an imagery
condition to test whether this also produces negative emotional
effects. As discussed, our original imagery condition may be
unlikely to provoke comparative thinking, but it could be adapted.
For example, our present instructions promote the use of field
perspective (first-person) imagery, but we predict that alternative
instructions to use observer perspective (third-person) imagery
would be more likely to facilitate comparative thinking (Kuyken &
Howell, 2006) and lose positive effects.
General Discussion
Our present experiments highlight differences between imagery
and verbal processing instructions for positive CBM-I. First, in
Experiment 1, we replicated the relative effectiveness of our pos-
itive imagery training condition on improving mood and bias,
extending this to a later test of emotional vulnerability (a
depression-relevant negative mood induction). That is, with iden-
tical positive training material, inferior effects on mood and bias
were obtained with our standard verbal instruction condition (Ex-
periment 1). Over the course of the training session, mood dete-
riorated in the verbal condition, whereas it improved within the
imagery condition (Experiment 1). The relatively negative effects
of the verbal condition may be due, at least in part, to participants
making unfavorable personal comparisons with the highly positive
material (Experiment 2). Overall, our results underscore the need
to study training instructions and information processing when
developing CBM. The inferiority of certain verbal conditions may
present a potential source of undesired effects when attempting
positive interpretation training, and possibly for other CBM para-
digms (see the introduction).
In addition to the literature previously reviewed, our present
findings are in line with research in the area of depressive rumi-
85
SPECIAL SECTION: POSITIVE BIAS TRAINING
nation. Drawing on Teasdale and Barnard (1993), Watkins (2004)
characterized a maladaptive form of ruminative processing as
“conceptual-evaluative,” which involves a more analytic focus on
the causes, meanings and consequences, including “thinking about
the self, focusing on discrepancies between current and wanted
outcomes” (p. 1039). Recent support for the distinction between
brooding rumination as “maladaptive” and reflection as “adaptive”
comes from Rude, Maestas, and Neff (2007), who found signifi-
cantly higher positive correlations between brooding, depression,
and anxiety than between reflection and these mood states. Making
unfavorable comparisons with positive information, as in our ex-
periment, may tap into one aspect of maladaptive rumination.
Interestingly and conversely, in Holmes and Mathews (2005), a
verbal negative CBM-I condition led to less negative affect than an
imagery negative CBM-I condition, suggesting that thinking ver-
bally helped reduce negative affect (cf. Sto¨ber & Borkovec, 2002).
In this experiment, the negative training scenarios were highly
negative, in comparison to which our average participant very
likely had more positive daily experiences. Therefore, downward
comparisons would be expected (Markman & McMullen, 2003)
with an associated more positive response to negative material.
However, rather than addressing rumination per se or responses
to thinking verbally about negative information, our research
agenda in the present article was to explore negative responses to
positive material. Our results are consistent with other depression
research indicating a difference in affective response to positive
information. Joormann and Siemer (2004) compared dysphoric
and nondysphoric participants’ response with thinking about pos-
itive autobiographical memories after inducing a negative mood.
Although the sadness ratings of nondysphoric participants im-
proved, those of dysphorics did not. Feldman, Joormann, and
Johnson (in press) discussed dampening responses as engaging in
thoughts that would likely shorten the duration of positive affect.
Self-report of dampening was associated with both rumination and
depression. It is possible that making unfavorable comparisons
with positive information contributes to dampening and the failure
of dysphorics to benefit from positive memories.
Future research using positive material within the CBM para-
digm could seek to reduce the salience of unfavorable compari-
sons. For example, Mathews et al. (2007) exposed participants to
positive training descriptions in a graded manner. Other methods
could include instructions not to compare or stimulus speeding, as
attempted in our experimental manipulation. Verbal processing
might be particularly conducive to engaging in comparative think-
ing due to rich semantic networks available in a verbal mode.
Switching to use imagery may be useful in this regard. Future
studies also need to fractionate the effects of making comparisons
and imagery more precisely. For example, observer perspective
imagery may facilitate comparisons (Kuyken & Howell, 2006) and
reduced affect. We predict that field imagery will be most bene-
ficial for positive CBM.
A limitation of the present experiments is that we only contrast
imagery versus verbal conditions. There are numerous control
conditions and questions yet to be explored. Future research should
use a third “control” condition to acquire baseline data concerning
the trajectory of mood across the session in the absence of either
training manipulation. Such a condition would allow the conclu-
sion that not only whether the original verbal and imagery condi-
tions differed in their emotional impact but also whether they made
a generally negative or a generally positive contribution. It would
clearly be clinically beneficial to develop CBM-I paradigms,
which are enjoyable rather than aversive to do. However, it may be
a challenge to design the appropriate control condition if it is to be
an alternative to imagery or verbal, because it is unclear what other
mode could be instructed. If participants are asked whether they
think in a way that seems like having neither mental images nor
verbal thoughts, they only infrequently endorse this, and if they do,
they find it hard to describe (Holmes, Mathews, Mackintosh, &
Dalgleish, in press). Alternatively, a “no-instruction” condition
could be argued to be a test of spontaneous tendency of how much
participants used imagery versus verbal, which would then need to
be assessed and statistically controlled for.
Future studies could also add a no-training control condition (in
which the positive training contingency is not present) to deter-
mine whether, in contrast, our positive training conditions exerted
any impact on emotion. An interesting possibility is that just
increasing imagery in the absence of a positive training contin-
gency may have benefits for depression (e.g., Moberly & Watkins,
2006). The autobiographical memory literature (as reviewed by
Williams et al., 2007) suggests that overgenerality (i.e., reduced
specificity) is associated with depression. We predict that promot-
ing imagery is an ideal candidate for increasing specificity, and
thus improving depressed mood. We also predict mood may be
further improved by combining an increased specificity bias with
a more positive interpretation bias. Thus, for CBM-I, a positive
imagery condition should exert a stronger impact than a no-
training (neutral) imagery control. Relatedly, negative imagery is
more powerful than neutral imagery CBM-I (Holmes & Mathews,
2005).
There are clearly many different ways in which imagery and
verbal conditions can differ. For example, imagery may encourage
more self-relevant processing (Holmes et al., in press), though in
the context of depressed mood, self-focus is generally considered
disadvantageous (Hertel & El-Messidi, 2006). Discovering what is
invited by imagery or verbal modes of processing offers research-
ers very interesting possibilities for future research. The field of
imagery and emotion is at a young stage, and we propose that an
exciting web of future research will be needed to delineate the
qualities and content of different types of imagery in relation to
emotion.
CBM paradigms are extremely useful in identifying active in-
gredients in processing difficulties underlying psychopathology
(Mathews & MacLeod, 2005). Furthermore, our clinically moti-
vated concern is to harness these paradigms to ameliorate process-
ing difficulties for treatment. As indicated by Experiment 1, our
positive imagery training condition may have benefits related to
both analogues of anxiety and depression, which may be useful in
reality given their high comorbidity (Moffitt et al., 2007). We need
to test whether beneficial effects of imagery CBM-I extend to a
clinical sample—that is, to determine whether the promise of our
CBM-I paradigms as a cognitive vaccine can translate to clinical
reality. A thrust of future research should be to improve the
positive imagery training condition and optimize real-world deliv-
ery. We suggest that harnessing mental imagery in interpretation
training and related CBM paradigms is a profitable future direc-
tion.
86 HOLMES, LANG, AND SHAH
References
Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New
York: International Universities Press.
Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck
Depression Inventory-II. San Antonio, TX: Psychological Corporation.
Bisson, M. A. S., & Sears, C. R. (2007). The effect of depressed mood on
the interpretation of ambiguity, with and without negative mood induc-
tion. Cognition and Emotion, 21, 614 – 645.
Brewin, C. R., & Holmes, E. A. (2003). Psychological theories of post-
traumatic stress disorder. Clinical Psychology Review, 23, 339 –376.
Butler, G., & Mathews, A. (1983). Cognitive processes in anxiety. Ad-
vances in Behaviour Research and Therapy, 5, 51– 62.
Cocude, M., Charlot, V., & Denis, M. (1997). Latency and duration of
visual mental images in normal and depressed subjects. Journal of
Mental Imagery, 21, 127–142.
Conway, M. A. (2001). Sensory-perceptual episodic memory and its con-
text: Autobiographical memory. Philosophical Transactions of the
Royal Society of London Series B-Biological Sciences, 356(1413),1375–
1384.
Feldman, G., Joormann, J., & Johnson, S. L. (in press). A self-report
measure of responses to positive affect: Rumination and dampening.
Cognitive Therapy and Research.
Fresco, D. M., Frankel, A. N., Mennin, D. S., Turk, C. L., & Heimberg,
R. G. (2002). Distinct and overlapping features of rumination and worry:
The relationship of cognitive production to negative affective states.
Cognitive Therapy and Research, 26, 179 –188.
Grey, S., & Mathews, A. (2000). Effects of training on interpretation of
emotional ambiguity. Quarterly Journal of Experimental Psychology
Section A: Human Experimental Psychology, 53, 1143–1162.
Hackmann, A., & Holmes, E. A. (2004). Reflecting on imagery: A clinical
perspective and overview of the special issue of memory on mental
imagery and memory in psychopathology. Memory, 12, 389 – 402.
Hayes, S. C., & Gifford, E. V. (1997). The trouble with language: Expe-
riential avoidance, rules and the nature of verbal events. Psychological
Science, 8, 170 –173.
Hertel, P. T. (2002). Cognitive biases in anxiety and depression: Introduc-
tion to the special issue. Cognition & Emotion, 16, 321–330.
Hertel, P. T., & El-Messidi, L. (2006). Am I blue? Depressed mood and the
consequences of self focus for the interpretation and recall of ambiguous
words. Behaviour Therapy, 37, 259 –268.
Higgins, E. T., Shah, J., & Friedman, R. (1997). Emotional responses to
goal attainment: Strength of regulatory focus as moderator. Journal of
Personality and Social Psychology, 72, 515–525.
Hirsch, C. R., Mathews, A., & Clark, D. M. (2007). Inducing an interpre-
tation bias changes self-imagery: A preliminary investigation. Behaviour
Research and Therapy, 45, 2173–2181.
Holmes, E. A., Crane, C., Fennell, M. J. V., & Williams, J. M. G. (2007).
Imagery about suicide in depression—“Flash-forwards”? Journal of
Behavior Therapy and Experimental Psychiatry, 38, 423– 434.
Holmes, E. A., & Mathews, A. (2005). Mental imagery and emotion: A
special relationship? Emotion, 5, 489 – 497.
Holmes, E. A., Mathews, A., Dalgleish, T., & Mackintosh, B. (2006).
Positive interpretation training: Effects of mental imagery versus verbal
training on positive mood. Behavior Therapy, 37, 237–247.
Holmes, E. A., Mathews, A., Mackintosh, B., & Dalgleish, T. (in press).
The impact of mental imagery on emotion assessed using picture-word
cues. Emotion.
Hyman, I. E., & Pentland, J. (1996). The role of mental imagery in the
creation of false childhood memories. Journal of Memory and Lan-
guage, 35, 101–117.
Joormann, J., & Siemer, M. (2004). Memory accessibility, mood regula-
tion, and dysphoria: Difficulties in repairing sad mood with happy
memories? Journal of Abnormal Psychology, 113, 179 –188.
Kavanagh, D., Andrade, J., & May, J. (2005). Imaginary relish and ex-
quisite torture: The elaborated intrusion theory of desire. Psychological
Review, 112, 446 – 467.
Kosslyn, S. M., Ganis, G., & Thompson, W. L. (2001). Neural foundations
of imagery. Nature Reviews: Neuroscience, 2, 635– 642.
Kuyken, W., & Brewin, C. R. (1994). Intrusive memories of childhood
abuse during depressive episodes. Behaviour Research and Therapy, 32,
525–528.
Kuyken, W., & Howell, R. (2006). Facets of autobiographical memory in
adolescents with major depressive disorder and never-depressed con-
trols. Cognition & Emotion, 20, 466 – 487.
Lau, M. A., Segal, Z. V., & Williams, J. M. G. (2004). Teasdale’s
differential activation hypothesis: Implications for mechanisms of de-
pressive relapse and suicidal behaviour. Behaviour Research and Ther-
apy, 42, 1001–1017.
Lawson, C., MacLeod, C., & Hammond, G. (2002). Interpretation revealed
in the blink of an eye: Depressive bias in the resolution of ambiguity.
Journal of Abnormal Psychology, 111, 321–328.
Mackintosh, B., Mathews, A., Yiend, J., Ridgeway, V., & Cook, E. (2006).
Induced biases in emotional interpretation influence stress vulnerability
and endure despite changes in context. Behavior Therapy, 37, 209 –222.
MacLeod, C., Rutherford, E., Campbell, L., Ebsworthy, G., & Holker, L.
(2002). Selective attention and emotional vulnerability: Assessing the
causal basis of their association through the experimental manipulation
of attentional bias. Journal of Abnormal Psychology, 111, 107–123.
Markman, K. D., Gavanski, I., Sherman, S. J., & McMullen, M. N. (1993).
The mental simulation of better and worse possible worlds. Journal of
Experimental Social Psychology, 29, 87–109.
Markman, K. D., & McMullen, M. N. (2003). A reflection and evaluation
model of comparative thinking. Personality and Social Psychology
Reviews, 7, 244 –267.
Marks, D. F. (1973). Visual imagery differences in the recall of pictures.
British Journal of Psychology, 64, 17–24.
Mathews, A., & Mackintosh, B. (2000). Induced emotional interpretation
bias and anxiety. Journal of Abnormal Psychology, 109, 602– 615.
Mathews, A., & MacLeod, C. (2002). Induced processing biases have
causal effects on anxiety. Cognition & Emotion, 16, 331–354.
Mathews, A., & MacLeod, C. (2005). Cognitive vulnerability to emotional
disorder. Annual Review of Clinical Psychology, 1, 167–195.
Mathews, A., Ridgeway, V., Cook, E., & Yiend, J. (2007). Inducing a
benign interpretational bias reduces trait anxiety. Journal of Behavior
Therapy and Experimental Psychiatry, 38, 225–236.
McLaughlin, K. A., Borkovec, T. D., & Sibrava, N. J. (2007). The effects
of worry and rumination on affect states and cognitive activity. Behavior
Therapy, 38, 23–38.
Moberly, N. J., & Watkins, E. (2006). Processing mode influences the
relationship between train rumination and emotional vulnerability. Be-
havior Therapy, 37, 281–291.
Moffitt, T. E., Harrington, H., Caspi, A., Kim-Cohen, J., Goldberg, D.,
Gregory, A. M., et al. (2007). Depression and generalized anxiety
disorder: Cumulative and sequential comorbidity in a birth cohort fol-
lowed prospectively to age 32 years. Archives of General Psychiatry, 64,
651– 660.
Morse, S., & Gergen, K. J. (1970). Social comparison, self-consistency,
and the concept of self. Journal of Personality and Social Psychology,
16, 148 –156.
Murphy, R., Hirsch, C. R., Mathews, A., Smith, K., & Clark, D. M. (2007).
Facilitating a benign interpretation bias in a high socially anxious
population. Behaviour Research and Therapy, 45, 1517–1529.
Pronin, E., & Wegner, D. M. (2006). Manic thinking: Independent effects
of thought speed and thought content on mood. Psychological Science,
17, 807– 813.
Reisberg, D., Pearson, D. G., & Kosslyn, S. M. (2003). Intuitions and
introspections about imagery: The role of imagery experience in shaping
87
SPECIAL SECTION: POSITIVE BIAS TRAINING
an investigator’s theoretical views. Applied Cognitive Psychology, 17,
147–160.
Rude, S., Maestas, K. L., & Neff, K. (2007). Paying attention to distress:
What’s wrong with rumination? Cognition & Emotion, 21, 843– 864.
Rude, S., Valdez, C. R., Odom, S., & Ebrahimi, A. (2003). Negative
cognitive biases predict subsequent depression. Cognitive Therapy and
Research, 27, 415– 429.
Salemink, E., van den Hout, M., & Kindt, M. (2007). Trained interpretive
bias and anxiety. Behaviour Research and Therapy, 45, 329 –340.
Schacter, D. L., Addis, D. R., & Buckner, R. L. (2007). Remembering the
past to imagine the future: The prospective brain. Nature Reviews:
Neuroscience, 8, 657– 661.
Spielberger, C. D., Gorsuch, R. L., Lushene, R., Vagg, P. R., & Jacobs,
G. A. (1983). Manual for State-Trait Anxiety Inventory. Palo Alto, CA:
Consulting Psychologists Press.
Sto¨ ber, J., & Borkovec, T. D. (2002). Reduced concreteness of worry in
generalized anxiety disorder: Findings from a therapy study. Cognitive
Therapy and Research, 26, 89 –96.
Strauman, T. J., & Higgins, E. T. (1987). Automatic activation of self-
discrepancies and emotional syndromes: When cognitive structures in-
fluence affect. Journal of Personality and Social Psychology, 53, 1004 –
1014.
Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics (3rd
ed.). New York: HarperCollins.
Teasdale, J. D., & Barnard, P. J. (1993). Affect, cognition and change:
Re-modelling depressive thought. Hove, England: Erlbaum.
Treynor, W., Gonzalez, R., & Nolen-Hoeksema, S. (2003). Rumination
reconsidered: A psychometric analysis. Cognitive Therapy and Re-
search, 27, 247–259.
Velten, E. (1968). A laboratory task of induction of mood states. Behaviour
Research and Therapy, 6, 473– 482.
Watkins, E. (2004). Adaptive and maladaptive ruminative self-focus during
emotional processing. Behaviour Research and Therapy, 42, 1037–
1052.
Watkins, E., & Moulds, M. (2005). Distinct modes of ruminative self-
focus: Impact of abstract versus concrete rumination on problem solving
in depression. Emotion, 5, 319 –328.
Watson, D., & Clark, L. A. (1994). The PANAS-X-Manual for the Positive
and Negative Affect Schedule-Expanded form. Unpublished manuscript,
University of Iowa.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Wenzlaff, R. M. (1993). The mental control of depression: Psychological
obstacles to emotional well-being. In D. M. Wegner & J. W. Pennebaker
(Eds.), Handbook of mental control (pp. 239 –257). Englewood Cliffs,
NJ: Prentice Hall.
Williams, J. M. G., Barnhofer, T., Crane, C., & Beck, A. T. (2005).
Problem solving deteriorates following mood challenge in formerly
depressed patients with a history of suicidal ideation. Journal of Abnor-
mal Psychology, 114, 421– 431.
Williams, J. M. G., Barnhofer, T., Crane, C., Hermans, D., Raes, F.,
Watkins, E., et al. (2007). Autobiographical memory specificity and
emotional disorder. Psychological Bulletin, 133, 122–148.
Williams, J. M. G., Ellis, N. C., Tyers, C., Healy, H., Rose, G., &
MacLeod, A. K. (1996). The specificity of autobiographical memory and
imageability of the future. Memory & Cognition, 24, 116 –125.
Wilson, E. J., MacLeod, C., Mathews, A., & Rutherford, E. M. (2006). The
causal role of interpretive bias in anxiety reactivity. Journal of Abnormal
Psychology, 115, 103–111.
Yiend, J., Mackintosh, B., & Mathews, A. (2005). The enduring conse-
quences of experimentally induced biases in interpretation. Behaviour
Research and Therapy, 43, 779 –797.
Received June 28, 2007
Revision received March 24, 2008
Accepted March 25, 2008 䡲
88 HOLMES, LANG, AND SHAH
