ArticlePDF Available

Threat-related attentional bias in anxious and nonanxious individuals: a meta-analytic study. [Meta-Analysis Research Support, Non-U.S. Gov't]


Abstract and Figures

This meta-analysis of 172 studies (N = 2,263 anxious,N = 1,768 nonanxious) examined the boundary conditions of threat-related attentional biases in anxiety. Overall, the results show that the bias is reliably demonstrated with different experimental paradigms and under a variety of experimental conditions, but that it is only an effect size of d = 0.45. Although processes requiring conscious perception of threat contribute to the bias, a significant bias is also observed with stimuli outside awareness. The bias is of comparable magnitude across different types of anxious populations (individuals with different clinical disorders, high-anxious nonclinical individuals, anxious children and adults) and is not observed in nonanxious individuals. Empirical and clinical implications as well as future directions for research are discussed.
Content may be subject to copyright.
Threat-Related Attentional Bias in Anxious and Nonanxious Individuals:
A Meta-Analytic Study
Yair Bar-Haim, Dominique Lamy, and
Lee Pergamin
Tel Aviv University
Marian J. Bakermans-Kranenburg and
Marinus H. van IJzendoorn
Leiden University
This meta-analysis of 172 studies (N 2,263 anxious, N 1,768 nonanxious) examined the boundary
conditions of threat-related attentional biases in anxiety. Overall, the results show that the bias is reliably
demonstrated with different experimental paradigms and under a variety of experimental conditions, but
that it is only an effect size of d 0.45. Although processes requiring conscious perception of threat
contribute to the bias, a significant bias is also observed with stimuli outside awareness. The bias is of
comparable magnitude across different types of anxious populations (individuals with different clinical
disorders, high-anxious nonclinical individuals, anxious children and adults) and is not observed in
nonanxious individuals. Empirical and clinical implications as well as future directions for research are
Keywords: attentional bias, selective attention, anxiety, threat
A normative function of the mechanisms underlying fear is to
facilitate detection of danger in the environment and to help the
organism respond effectively to threatening situations. Biases in
processing threat-related information have been assigned a prom-
inent role in the etiology and maintenance of anxiety disorders
(A. T. Beck, 1976; Eysenck, 1992; Mathews, 1990; Mathews &
MacLeod, 2002; Williams, Watts, MacLeod, & Mathews, 1988).
Specifically, several authors have suggested that the attentional
system of anxious individuals may be distinctively sensitive to and
biased in favor of threat-related stimuli in the environment. Over
the last 2 decades, this notion has fostered intensive research on
attentional biases in anxiety using different experimental tasks
both in clinical populations displaying a variety of anxiety disor-
ders and in nonclinical individuals reporting high levels of anxiety.
Although several narrative reviews have been published on
topics related to processing biases in anxiety, they were typically
selective in nature. Some reviews have focused on processing
biases in selected populations, such as in individuals experiencing
posttraumatic stress disorder (PTSD; Buckley, Blanchard, & Neill,
2000), social phobia (Clark & McManus, 2002; Heinrichs &
Hofman, 2001; Musa & Lepine, 2000), obsessive–compulsive
disorder (OCD; Summerfeldt & Endler, 1998), generalized anxiety
disorder (GAD; Mogg & Bradley, 2005), or panic disorder and
phobias (McNally, 1999), or in children (Ehrenreich & Gross,
2002). Other reviews have focused on specific issues within the
field, such as processing biases in anxiety versus depression (Dal-
gleish & Watts, 1990; Mineka & Gilboa, 1998), nonverbal infor-
mation processing in anxiety (Mogg & Bradley, 2003), or process-
ing biases in anxiety as measured specifically by the emotional
Stroop paradigm (MacLeod, 1991; Williams, Mathews, & MacLeod,
1996). Other work has revolved on the literature relevant to testing
a specific model of threat processing in anxiety, such as Mogg and
Bradley’s (1999b) cognitive-motivational model, Fox’s (2004)
review on maintenance versus capture of attention in anxiety-
related biases, or Mansell’s (2000) top-down model of processing
biases in anxiety. The field lacks, however, an updated encom-
passing and systematic review. In addition, there has not been a
single quantitative review of the considerable amount of available
data assessing the overall effect size and the influence of the major
moderators related to threat processing in anxiety. For in-depth
discussions of the advantages inherent in meta-analysis versus
narrative reviews, see Cooper and Hedges (1994) and Cooper and
Lindsay (1998). Therefore, the objective of the present meta-
analysis was to assess the magnitude and boundary conditions of
threat-related biases in anxiety by organizing the extant database
according to variables identified as potential modulators of the
Theories of Cognitive Biases in Anxiety
Cognitive accounts of anxiety differ with regard to the roles they
assign to biases in attention, interpretation, memory, and judgment
in the etiology and maintenance of anxiety. According to schema
theories (e.g., A. T. Beck, 1976; A. T. Beck & Clark, 1997; A. T.
Beck, Emery, & Greenberg, 1985; Bower, 1981, 1987) cognitive
processing is guided by schemas that largely determine how in-
formation is attended to, interpreted, and remembered. In anxious
individuals, schemas are thought to be biased toward threat. As a
result, threat-related material is favored at all stages of processing,
Yair Bar-Haim, Dominique Lamy, and Lee Pergamin, the Adler Center
for Reasearch in Child Development and Psychopathology, Department of
Psychology, Tel Aviv University, Tel Aviv, Israel; Marian J. Bakermans-
Kranenburg and Marinus H. van IJzendoorn, Center for Child and Family
Studies, Leiden University, Leiden, the Netherlands.
Preparation of this article was supported by Israeli Science Foundation
Grant 989/03. Marinus H. van IJzendoorn was supported by the NWO/
Spinoza prize of the Netherlands Organization for Scientific Research.
Correspondence concerning this article should be addressed to Yair
Bar-Haim, Department of Psychology, Tel Aviv University, Ramat Aviv,
Tel Aviv 69978, Israel. E-mail:
Psychological Bulletin Copyright 2007 by the American Psychological Association
2007, Vol. 133, No. 1, 1–24 0033-2909/07/$12.00 DOI: 10.1037/0033-2909.133.1.1
including early processes such as attention and stimulus encoding
and later processes such as memory and interpretation.
Other theories suggest that anxious individuals are prone to
biases at specific stages of information processing. Some authors
have proposed that the attentional system of anxious individuals is
abnormally sensitive to threat-related stimuli and that these indi-
viduals tend to direct their attention toward threatening informa-
tion during early, automatic stages of processing (Williams et al.,
1988). This idea is consistent with the literature showing that
evaluation of stimulus emotional valence may take place at a very
early stage of processing, automatically, and in the absence of
awareness (e.g., LeDoux, 1995, 1996; Ohman, 1993). Abnormal-
ities in the threat-detection mechanism of anxious individuals
would therefore lead them to adopt a hypervigilant mode toward
threat. A quite different proposal is that inhibition of detailed
processing of threatening information is the core deficit in anxiety,
which is reflected in avoidance of threatening stimuli (Foa &
Kozak, 1986; Mogg, Bradley, De Bono, & Painter, 1997). Accord-
ing to this view, threat-related biases in anxiety are confined to
later stages of processing.
More recent models have suggested a more complex pattern of
biases that may reconcile these apparently conflicting views. They
have emphasized the time course of attentional allocation in main-
taining high levels of anxiety (e.g., Clark & Wells, 1995; Eysenck,
1992; Mathews & Mackintosh, 1998; Mogg & Bradley, 1998;
Williams, Watts, MacLeod, & Mathews, 1997).
Williams et al. (1997, 1988) and others (see Amir, Foa, & Coles,
1998; Mogg et al., 1997) have proposed that anxious individuals
tend to direct their attention toward threat during early, automatic
stages of processing, whereas during later, more strategic stages of
processing, they tend to direct their attention away from threat.
Automatic allocation of attentional resources to threat-related
stimuli would serve to enhance an individual’s anxious state, yet
subsequent avoidance of such stimuli would prevent more elabo-
rate evaluation processes that could deflate the threatening value of
these stimuli and thereby reduce anxiety.
Other research groups (e.g., Fox, Russo, Bowles, & Dutton,
2001; Fox, Russo, & Dutton, 2002; Yiend & Mathews, 2001) have
suggested that anxiety has little impact on initial detection of threat
but has a stronger effect in modulating the maintenance of atten-
tion on the source of threat. That is, they have proposed that a
delay in disengaging from threat stimuli might be the primary
attentional difference between anxious and nonanxious individu-
Despite disagreements as to the specific cognitive mechanisms
underlying anxiety, there is a consensus at the theoretical level that
anxiety is associated with biases in attending to threat-related
information. However, at the empirical level, the wide range of
studies concerned with threat-related biases in anxiety offers a
somewhat confusing picture, plagued by contradictory findings
that lack unambiguous explanation. In the next section, we de-
scribe potential factors contributing to the divergence of empirical
findings and their interpretation in the field.
Operational Considerations in the Study of Attentional
Bias in Anxiety
Part of the confusion regarding the empirical findings may be
traceable to the fact that two different operational definitions of
bias are used in the literature but do not necessarily converge. The
first definition refers to a significant difference in the attentional
allocation of highly anxious individuals with respect to threat-
related stimuli relative to neutral stimuli (a within-subject bias).
The other definition refers to a significant difference between
anxious and nonanxious individuals in the pattern of attentional
allocation to threat-related and neutral stimuli (a between-subjects
Anxious individuals are sometimes found to attend preferen-
tially to threat-related stimuli relative to neutral stimuli (within-
subject bias) but not significantly more so than control participants
(e.g., Kyrios & Iob, 1998). Such findings imply a threat-related
bias that does not specifically characterize anxious individuals (no
between-subjects bias). Conversely, a significant difference in the
attentional allocation pattern of anxious individuals relative to
nonanxious controls (between-subjects bias) is sometimes re-
ported, while at the same time, anxious individuals show no
significant tendency to allocate their attention to threatening rela-
tive to neutral material (no within-subject bias). This pattern of
results may arise, for example, when anxious participants show a
nonsignificant attentional bias toward threat whereas control par-
ticipants show a bias away from threat (e.g., Musa, Lepine, Clark,
Mansell, & Ehlers, 2003; Stewart, Conrod, Gignac, & Pihl, 1998).
Yet, in both instances of divergence between the within-subject
and between-subjects definitions of the bias, the finding that is
consistent with the idea of anxiety’s being associated with a
threat-related attentional bias has most often been put forward,
which contributes to the impression that the threat-related bias in
anxious individuals is large and robust.
In addition, studies of the threat-related bias in anxiety usually
differ along a wide array of variables, with positive findings of an
attentional bias often being found in one of the relevant experi-
mental conditions but not in others within the same study. The
myriad of variables used in studies of threat-related biases in
anxiety can be grouped under two broad categories: procedural
variables and population-related variables.
With regard to procedural variables, the threat-related bias has
been measured using different paradigms, namely, the emotional
Stroop, dot-probe, emotional spatial cuing, and visual search par-
adigms. Although there is strong evidence that all these paradigms
reflect the operation of attentional processes (e.g., Driver, 2001), it
is also generally accepted that they do not tap the same aspects of
attention (e.g., Shalev & Algom, 2000). In addition, studies have
differed in the types of stimuli used, either printed words or
pictorial stimuli. Finally, the threat-related bias has been examined
in conditions that prevented conscious perception (subliminal ex-
posure) and in conditions that allowed clear awareness (supralim-
inal exposure). Processes triggered by signs of threat that are
perceived with or without awareness might yield different behav-
ioral outcomes. In addition, with consciously perceived stimuli,
attentional effects have been measured at various times after
stimulus presentation. Given the debate concerning the notions of
hypervigilance versus avoidance of threat in anxiety, the time
course of attentional allocation to threat-related stimuli is impor-
tant, because different time frames are likely to yield different
With regard to population-related variables, anxious participants
have been sampled from populations that varied considerably from
one study to another. Some studies have examined adults, whereas
others have examined children. Participants either were diagnosed
with clinical anxiety (clinical population) or only scored high on
questionnaires relying on self-report of either state or trait anxiety
(nonclinical population). Moreover, clinically diagnosed partici-
pants differed in the type of anxiety disorder they experienced,
namely, GAD, specific phobias, social phobia, OCD, PTSD, and
panic disorder. Whether these participants concomitantly experi-
enced depression was often reported yet seldom controlled for.
Thus, in view of the numerous experiments with diverging
results, a quantitative test of the overarching conclusions typically
drawn as to the existence of a threat-related bias in anxious
individuals is in order. We now turn to a more detailed description
of the different moderators considered in this meta-analysis.
Procedural Moderators
Subliminal Versus Supraliminal Stimulus Presentation
Research on the neural substrates of emotion has underscored
the role of automatic processes in mediating anxiety and fear
responses (e.g., LeDoux, 1996; Ohman, 1993). Specifically, it has
been suggested that neural structures sensitive to signs of biolog-
ically threat-relevant stimuli can directly trigger anxiety responses
and draw attention toward the source of threat before conscious
perception and evaluation occur. Most studies of the threat-related
bias in anxiety have involved clearly visible stimuli, typically
presented for 500 ms or longer, an exposure time that allows the
stimuli to be consciously perceived. However, in a number of
studies the critical stimuli were presented in conditions that pre-
cluded conscious processing. In these studies, the stimuli were
typically presented for very brief durations (subliminal exposure)
and were immediately masked. Such backward masking is known
to interrupt sensory processing, thereby preventing the masked
stimuli from reaching awareness (e.g., Di Lollo, Enns, & Rensink,
2000). The rationale for using subliminal stimuli is that the struc-
tures underlying early, automatic, rather than later, conscious
processing of threat may be abnormally sensitive in anxious indi-
Finding a threat-related bias in response to supraliminal stimuli
does not allow a distinction between the contributions of a pre-
conscious bias versus a bias that requires awareness of the threat-
ening stimulus. Finding a bias using subliminal stimuli can be
accounted for only by an early, preconscious bias.
Experimental Paradigms
Three main experimental paradigms have been used to study the
threat-related attentional bias in anxiety: emotional Stroop, dot-
probe, and emotional spatial cuing. In the following paragraphs,
we describe these paradigms and the different theoretical implica-
tions that arise from the use of each of them. Surprisingly, the
visual search paradigm, a primary tool used to investigate atten-
tional priority when several objects compete for attention, has been
only seldom used in the context of anxiety research (cf. Gilboa
Schechtman, Foa, & Amir, 1999; Hadwin et al., 2003; Rinck,
Becker, Kellermann, & Roth, 2003). Thus, a meta-analysis of these
studies would be premature.
The emotional Stroop is a modified version of the classic
color-naming Stroop interference paradigm (Stroop, 1935). The
Stroop effect refers to the difference in color-naming performance
between congruent (e.g., the word red printed in red) and incon-
gruent (e.g., the word red printed in green) stimuli. The presence
of the Stroop effect documents the failure to focus exclusively on
the target dimension of color. In the emotional Stroop, the word
valence instead of its semantic congruence with the printed color
is manipulated. For instance, response latency to name the printed
color of a word is compared when this word is threat related (e.g.,
cancer”) relative to when it is neutral (e.g., “plate”). When
pictures instead of words are used, the participant might be re-
quired to name the color of a schematic face, with this face
displaying either a neutral or an angry expression. Threat-related
bias is inferred when color naming takes longer with a threat
stimulus relative to a neutral stimulus (MacLeod, 1991).
The emotional Stroop was initially the most widely used tool to
investigate threat-related attentional biases in anxiety. However, it
has been criticized with the argument that delayed response laten-
cies with threat-related stimuli may result from late processes that
are unrelated to attention (e.g., Algom, Chajut, & Lev, 2004;
MacLeod, Mathews, & Tata, 1986). MacLeod et al. (1986) sug-
gested that anxious participants might process both the neutral and
the threat-related meanings to the same degree but that the pres-
ence of the latter might intensify the negative affective state of
anxious participants to a level where it impairs reaction time. De
Ruiter and Brosschot (1994) further suggested that interference by
threat stimuli in the emotional Stroop might reflect effortful avoid-
ance of processing threat cues rather than attentional capture by
these cues.
To overcome these problems MacLeod et al. (1986) designed
the dot-probe paradigm. In this task, two stimuli, one threat-
related and one neutral, are shown briefly on each trial, and their
offset is followed by a small probe in the location just occupied by
one of them. Participants are required to respond as fast as possible
to the probe. On the basis of the attention literature (e.g., Navon &
Margalit, 1983), response latencies on the dot-probe task are held
to provide a “snapshot” of the distribution of participants’ atten-
tion, with faster responses to probes presented in the attended
relative to the unattended location. Attentional bias toward threat
is revealed when participants are faster to respond to probes that
replace threat-related rather than neutral stimuli.
In the dot-probe paradigm, participants are required to respond
to a neutral stimulus (the probe). Therefore, there is no concern
that delayed latencies may result from response bias or general
arousal. An additional advantage of this paradigm is that manip-
ulating stimulus onset asynchrony, that is, the time interval be-
tween presentation of the critical stimuli and presentation of the
probe, allows for investigating the time course of attentional allo-
In the dot-probe paradigm, the advantage in performance on
trials in which the target probe appears at the location of the
threat-related stimulus might result either from faster engagement
with the threat stimulus or from a difficulty to disengage from it.
In order to determine the relative contributions of these two
components of attention (Posner & Peterson, 1990) to the threat-
related attentional bias, a variant of Posner’s spatial cuing para-
digm has been used. In Posner’s classical paradigm (Posner, 1980),
a cue appears in one of two locations and is followed by a target
presented at the cued location on a majority of the trials (valid-cue
condition) and at the alternative location on a minority of the trials
(invalid-cue condition). Performance in detecting or identifying
the target is typically faster on validly cued than on invalidly cued
trials. Speeding on validly cued trials has been attributed to the
benefits of attentional engagement with the cued location. Slowing
on invalidly cued trials has been associated with the costs of
having to disengage attention from the cued location.
Stormark, Nordby, and Hugdahl (1995) adapted this paradigm
to the study of attentional allocation to emotionally valenced
stimuli by manipulating the emotional content of the cue. Fox et al.
(2001) later used the emotional spatial cuing paradigm to inves-
tigate attentional bias related to threat in anxiety. A threat-related
attentional bias is revealed when validity effects (performance on
invalid-cue trials minus performance on valid-cue trials) are larger
when the cue is threat related than when it has a neutral content.
Moreover, a valence-related modulation of performance on valid-
cue trials indicates that the attentional bias occurs at the stage of
initial orienting of attention, whereas such modulation on invalid-
cue trials reflects a difficulty in disengaging attention from threat-
related material after attention has been engaged. Unfortunately,
the small number of studies that have distinguished between the
“engage” and “disengage” components of attention, and the fact
that these components were invariably examined in within-subject
designs, precluded meta-analytic examination of this issue.
As with the dot probe, in the emotional spatial cuing paradigm
the target to which subjects respond is neutral with respect to
valence. Thus, this paradigm is not open to the alternative account
of response bias in anxious participants. However, unlike the
dot-probe paradigm, in which the neutral and threat-related stimuli
compete for the participants’ attention, in the emotional spatial
cuing paradigm the visual field contains just one stimulus. To the
extent that competition among different stimuli might be a prereq-
uisite for threat-related attentional bias in anxiety to emerge, the
dot-probe task might be a more sensitive paradigm. Moreover,
whereas in the dot-probe paradigm, the valenced cue stimuli are
utterly task irrelevant, in the emotional spatial cuing paradigm
participants are instructed to attend to the valenced cue (through
validity instructions). This fact may hamper the generalizability of
the findings generated by the emotional spatial cuing paradigm, as
these might be contingent on the cue stimulus’s being task rele-
Words Versus Naturalistic Stimuli
Threat-related bias in anxiety has been initially investigated
using word stimuli, the meaning of which was either threat related
or neutral. Reliance on verbal stimuli was criticized on several
grounds (e.g., Bradley et al., 1997). It is likely that anxious
individuals spend more time than nonanxious individuals thinking
about threatening events or speaking with others about their anx-
ious feelings. As a result, the very words used as threat stimuli are
likely to be more primed in anxious participants than in nonanx-
ious control participants. Thus, the valence-related effects ob-
served with words may reflect high familiarity and subjective
frequency of use of the threat words rather than a threat-related
attentional bias (McNally, Riemann, & Kim, 1990).
Threat-related and neutral pictures have been used to corrobo-
rate the findings obtained with word stimuli. The most widely used
pictorial stimuli are human faces displaying emotion expressions.
Recognition of facial expression is automatic and does not require
conscious awareness (e.g., Morris, Ohman, & Dolan, 1998). Such
rapid recognition is highly adaptive: A threatening face is a clear
sign of danger and should therefore be a good candidate for
capturing the attention of anxious individuals. Angry and fearful
faces fall into the category of threat-related stimuli, whereas neu-
tral or happy faces are typically used as control stimuli.
Population-Related Variables in the Study of Threat-
Related Bias in Anxiety
Clinically Diagnosed Anxiety Versus Nonclinical Self-
Reported Anxiety
The attentional bias in anxiety has been investigated in individ-
uals with diagnosed anxiety disorders (clinical population) and in
individuals who self-reported high levels of anxiety on question-
naires (nonclinical population). As clinical participants typically
display more severe anxiety, it is reasonable to expect that they
will show larger attentional biases. In addition, some authors have
claimed that even with similar levels of reported anxiety, clinical
participants should display a larger bias due to qualitative differ-
ences between clinical and nonclinical anxiety (e.g., Martin, Wil-
liams, & Clark, 1991).
One should keep in mind, however, that in between-subjects
analyses (anxious vs. control) for clinical versus nonclinical pop-
ulations, control participants for the nonclinical populations are
often selected for particularly low anxiety scores, whereas controls
for clinical participants are a matched sample, usually selected
from the general population.
Threat-related biases have been investigated in a variety of
clinical anxiety disorders (e.g., GAD, PTSD, social phobia, simple
phobias, OCD, panic disorder). Although the different disorders
present considerable differences in symptoms, time course, etiol-
ogy, and prognosis, they are typically held to belong to the same
overarching family of anxiety disorders. Finding that the threat-
related attentional bias is reliably detected in each of these disor-
ders would reinforce the idea that attentional bias is a core com-
ponent of anxiety. In contrast, if the attentional bias is not reliably
demonstrated in one or more of the disorders, this may be inter-
preted as a divergence of this particular disorder from the family of
anxiety disorders (see the discussion regarding OCD in Summer-
feldt & Endler, 1998).
Effect of Depression Comorbidity
Anxiety and depression are frequently comorbid conditions
(e.g., Mineka, Watson, & Clark, 1998). Despite common features
such as high levels of negative affect and distress, the two condi-
tions also present unique features such as exacerbated fear of
danger in anxiety and thoughts of failure and worthlessness in
depression. With regard to cognitive biases, it has been suggested
that anxiety is associated with an attentional bias toward threat and
depression is associated with greater elaboration of negative ma-
terial (Williams et al., 1997). Accordingly, patients with depres-
sion do not generally show an attentional bias toward negative
information and appear to do so only with long stimuli exposures
that allow later, elaborative processing to occur (Eizenman et al.,
2003; see Mogg & Bradley, 2005, for an extensive review).
It is not clear at this point how anxiety and depression interact
on measures of attentional bias. On the one hand, if depression-
related effects are revealed only at later stages of processing, then
comorbidity of depression might not affect the attentional bias in
anxiety. On the other hand, there have been reports that when
clinical anxiety and depression coexist, the attentional bias is no
longer found (e.g., Mogg, Bradley, Williams, & Mathews, 1993).
Mogg, Bradley, and Williams (1995) suggested that the attentional
bias in anxiety may reflect an individual’s readiness to orient
toward negative stimuli in order to deal with potential threats, that
is, a motivational state. They further reasoned that depression is an
amotivational state, and that depression comorbidity might there-
fore inhibit the motivation-based selectivity that characterizes anx-
iety. Thus, assessing the potential effects of participants’ depres-
sion comorbidity may be of crucial importance for studies of
threat-related bias in anxiety. In practice, however, many studies
have failed to either report participants’ depression levels or to
covary them out from the attentional bias effect.
State Versus Trait Anxiety
Threat-related attentional bias in anxiety has been investigated
mainly for trait anxiety and more seldom for state anxiety. Very
few studies have directly compared the effects of state versus trait
anxiety on the attentional bias. Some of these have examined
which, of state or trait anxiety scores, were best correlated with
attentional bias scores in clinically anxious individuals (e.g.,
Mathews & MacLeod, 1985; Mogg, Mathews, & Weinman, 1989).
Others have used nonclinical participants differing in levels of trait
anxiety and either experimentally induced state anxiety (e.g., Rich-
ards, French, Johnson, Naparstek, & Williams, 1992) or took
advantage of naturally occurring stressful events (e.g., MacLeod &
Rutherford, 1992). These studies have yielded conflicting findings
and have led to various proposals as to the relative roles of state
and trait anxiety in the attentional bias. For instance, Broadbent
and Broadbent (1988) suggested that the two factors interact, with
the effect of state anxiety being substantially larger in individuals
with high trait anxiety than in individuals with low trait anxiety. A
somewhat different proposal is that whereas state anxiety and the
attentional bias are positively correlated in individuals with high
trait anxiety, they are negatively correlated in individuals with low
trait anxiety (Egloff & Hock, 2001). Others suggested that both
transient stress (irrespective of trait anxiety) and enduring anxious
personality characteristics (irrespective of state anxiety) are suffi-
cient to produce the attentional bias (Mogg, Mathews, Bird, &
Macgregor-Morris, 1990). Thus, how state and trait anxiety con-
tribute to threat-related attentional bias remains an unresolved
issue. In the present meta-analysis we compared the effect sizes of
studies using state versus trait anxiety as the grouping variable of
anxiety, but we did not examine the interaction between the two
factors because of a lack of studies directly tackling this issue.
Children Versus Adult Participants
Because anxiety is highly prevalent in children and adolescents
(Albano, Chorpita, & Barlow, 2003), and because continuity of
trait anxiety from childhood to adulthood has been widely docu-
mented (e.g., Feehan, McGee, & Williams, 1993; Ferdinand &
Verhulst, 1995; Pine, Cohen, Gurley, Brook, & Ma, 1998), there is
clear theoretical and practical significance to the study of devel-
opmental aspects of the threat-related bias in anxiety. Surprisingly,
however, only a small number of studies have been published with
children relative to the extensive research with adults.
The conflicting findings in the extant literature have given rise
to different proposals as to the developmental course of the asso-
ciation between processing biases and anxiety. For instance, Kindt
and colleagues (Kindt, Bierman, & Brosschot, 1997; Kindt,
Brosschot, & Everaerd, 1997) have suggested that both anxious
and nonanxious young children show equal bias with respect to
threat-related stimuli, and that whereas nonanxious children learn
to inhibit this bias with increasing age, anxious children do not.
According to this view, a significant attentional bias is expected in
both anxious and nonanxious young children (Kindt & van den
Haut, 2001). Others (e.g., Martin, Horder, & Jones, 1992) have
proposed that threat-related attentional biases and anxiety go hand
in hand from very early on in life. Because of the relatively small
number of studies and the lack of longitudinal research, the present
meta-analysis only allowed us to compare the effect sizes of
studies with children versus adults on some of the studied mod-
With the aforementioned considerations in mind, the purposes
of the present meta-analysis were as follows: First, we wanted to
assess the overall effect size of the attentional bias in anxiety in the
literature to date and to determine whether the bias is smaller,
absent, or away from threat in nonanxious participants. Second, we
wanted to provide a systematic evaluation of the effects of the
major variables along which the existing studies on attentional bias
in anxiety differ. With respect to procedural variables, we exam-
ined whether subliminal stimuli exposure times result in smaller
effect sizes than supraliminal stimuli exposure times, whether or
not divergent results are found for the various paradigms, and
whether naturalistic pictures produce a stronger bias than word
stimuli. With respect to population-related variables, we examined
whether clinically anxious participants display a larger bias than
high-anxious participants recruited from normative populations,
whether or not the bias is characteristic of all anxiety disorders,
and whether effect sizes for threat-related bias in anxious children
and adolescents are similar to those in anxious adults. The third
purpose of the present study was to use the meta-analytic tool to
generate new findings by analyzing interactions among the differ-
ent moderators, whether procedural or population related.
Literature Base
Studies were collected through a search of the PsycInfo and PubMed
databases using the key words attention*, bias*, selective attention*,
Stroop, dot-probe, probe detection, Posner, spatial cueing, or visual
search, intersected with anxi* (anxiety), phob* (phobia), PTSD, OCD,
panic,orGAD. In addition, these databases were searched with the names
of researchers in the field to see whether there were additional relevant
publications from these authors. The references of all the obtained articles
as well as of relevant review articles (e.g., Dalgleish & Watts, 1990; Mogg
& Bradley, 1999b; Williams et al., 1996) were systematically searched for
additional relevant studies.
Inclusion Criteria
We used the following criteria to select studies for inclusion in the
1. The study was published as a journal article in the English
language until May 2005.
2. The study used one of the following experimental paradigms:
emotional Stroop, probe detection (or dot-probe) task, or a ver-
sion of the emotional spatial cuing task. There were insufficient
visual search studies meeting the inclusion criteria as outlined
later to allow systematic analysis.
3. The difference between threat-related and neutral stimuli could
be assessed. Studies that compared threat-related stimuli with
stimuli other than neutral (e.g., Mathews & MacLeod, 1985, who
contrasted reactions to negative stimuli with a combination of
positive and neutral stimuli) were excluded from the meta-
analysis. Differences between threat-related stimuli and other
emotionally valenced stimuli render the source of the bias un-
clear, because the observed behavior could stem from a bias
related to the threat stimulus or from a bias related to the other
emotionally valenced stimulus, or from a combination of both. It
should be noted that this selection criterion precludes a distinc-
tion between the effects of threat in particular versus emotion in
general. The investigation of that distinction calls for a separate
set of meta-analyses comparing neutral versus emotional non-
threat stimuli, which is beyond the scope of the present study.
Similarly, studies comparing threat-related words to nonwords
(e.g., McNally, Riemann, Louro, Lukach, & Kim, 1992) were
excluded because an observed bias in such studies might stem
from a difference in semantic versus nonsemantic processing
rather than from the valence of the threat word.
4. The study included a group of anxious participants selected on
the basis of either clinical diagnosis or self-reported high anxiety
on a questionnaire. Studies that formed anxiety groups based
solely on experimental manipulations inducing anxiety in nor-
mative samples (e.g., Green, Rogers, & Elliman, 1995) were
excluded because the anxiety-inducing procedures differed con-
siderably between studies and were typically not validated by
reliable manipulation checks.
5. The study reported data that allowed the computation of an effect
size for at least one of the following outcome measures of
attentional bias: a within-group comparison for anxious partici-
pants, a within-group comparison for control participants, or a
between-groups comparison. The within-group effects refer to
the bias measured as the difference between threat-related and
neutral conditions reported by statistics such as t or F values. The
between-groups effect refers to the bias measured as the differ-
ence between the anxious and control groups on the aforemen-
tioned effect, reported with between-groups statistics such as t or
F values, or sometimes as a correlation between the bias score
and the anxiety score on a questionnaire.
6. In cases in which an effect was reported as nonsignificant but
exact statistics were not provided, we calculated an estimated
effect size assuming p .50, in order to ensure a representative
sample of outcomes (Cooper & Hedges, 1994). This procedure
was used only when enough information was provided (e.g.,
tables, figures) to determine the direction of the effect, and it was
used in 25.6% of the between-groups outcomes, in 17.9% of the
within-anxious-group outcomes, and in 40.2% of the within-
control-group outcomes.
The listed criteria resulted in the selection of 172 studies (including 4,031
participants in total), published between February 1986 and May 2005.
Coding System and Coding Decisions
We used a standard coding system to rate every study (see Table 1). We
coded sample size (N), whether the participants were children under 18
years of age or adults (age group), and whether the anxious group was
clinically diagnosed or included participants with high self-reported anxi-
ety (anxiety type). For clinical samples we noted the type of anxiety
disorder, and for self-reported anxiety, we noted the type of anxiety
measure used. We coded whether participants with comorbid mood disor-
der were included or excluded from each sample (mood disorder comor-
We coded the type of paradigm used (i.e., emotional Stroop, dot probe,
or emotional spatial cuing) and the type of stimuli presented in the
experiment (stimulus type; i.e., words or pictures). We also took into
account whether the stimuli in the emotional Stroop paradigm were pre-
sented in a blocked design (i.e., stimuli from different categories presented
within different blocks of trials) or randomly (emotional Stroop design).
Finally, to assess the effect of stimulus exposure time, we coded whether
the stimuli were presented with subliminal or supraliminal exposure. Be-
cause several dot-probe studies were theoretically concerned with the
effects of various supraliminal exposure times, we created a more detailed
exposure time moderator variable for dot-probe studies (subliminal, 500
ms, and 1,000 ms or longer).
Some additional coding decisions were made:
1. To enhance the power of moderator analyses, in studies in which
participants were tested on more than one level of a moderator
variable (e.g., participants were presented with both word and
picture stimuli), we selected the level that included fewer overall
samples for analysis. Specifically, we selected subliminal over
supraliminal exposure times, state over trait anxiety, the dot-
probe paradigm over the emotional Stroop paradigm, and picture
stimuli over word stimuli. In dot-probe studies not reporting
subliminal exposure times, we preferred longer exposures (1,000
ms or longer) over exposures of 500 ms.
2. When data for the same participants were reported in two dif-
ferent studies, only one study was selected based on the criteria
just mentioned.
3. If a single control group was compared with two or more anxiety
groups within the same study, the number of control participants
in the reported analyses was split accordingly to avoid inflation
of the number of participants.
4. When a study assigned participants to more than two groups of
anxiety level (e.g., low, medium, and high anxiety), we selected
the two extreme groups.
5. When different intensity levels of threat stimuli were used in one
study (e.g., mild threat and high threat), we calculated an effect
size that reflected the average of the two conditions. Although
different theories postulate different assumptions concerning the
effect of threat level on the magnitude and direction of bias in
anxious and in control participants (for a discussion on this topic,
see Mogg, McNamara, et al., 2000; Wilson & MacLeod, 2003),
most studies have used only one level of threat, and levels of
threat are hardly comparable across studies. Thus, when mild-
and high-threat stimuli are used in the same study, it is practically
impossible to determine which of the two threat levels is more
compatible with the “‘single”‘ level of threat stimuli used in
other studies. For these reasons, we chose to average the effects
of different intensities of threat instead of choosing one over the
6. When more than one type of threat-related stimuli were pre-
sented (e.g., physical threat and social threat) the type of threat
most congruent to the studied anxiety group was selected (e.g.,
for participants with social phobia, we selected the social threat
data over the physical threat data). In cases in which two threat
categories were equally relevant to the studied group of partici-
pants, we calculated an effect size that reflected the average of
the two effects. For example, Brosschot, de Ruiter, and Kindt
(1999) studied the threat-related bias in high- and low-trait-
anxious participants, who were presented with social threat
words and physical threat words, both equally relevant to the
general concept of trait anxiety. Thus, the average of the two
effect sizes was computed for further analyses.
7. When studies involved therapy or any kind of experimental
manipulation and measured selective attention to threat before
and after the manipulation, we included the data emanating from
the pretest measurement.
Intercoder reliability for the moderator variables was established on 15%
of the articles included in the meta-analysis. Across the total variable
matrix, kappas ranged from .69 to 1.00 (the mean kappa was .96). Dis-
agreements were resolved by discussion, and the final coding reflected the
consensus between the two coders.
Meta-Analytic Procedures
The effect size index we used for all outcome measures in the present
meta-analysis is Cohen’s d, that is, the difference between the means of
two conditions or groups divided by their pooled standard deviation. For
the within-group analyses, a positive sign of the effect size value indicates
that attention is biased toward threat-related stimuli. For the between-
groups analyses, a positive sign of d indicates that the attentional bias
toward threat is larger in anxious participants than in control participants.
All analyses and computations were carried out using Comprehensive
Meta-Analysis software, Version 2.002 (Biostat, Englewood, NJ).
Because most of our data sets were heterogeneous in their effect sizes,
and because random effects models are somewhat more conservative than
fixed effects parameters in such cases, combined effect sizes and their
confidence intervals (CIs) are presented in the context of random effects
models. For a detailed discussion of this issue, see Bakermans-Kranenburg,
van IJzendoorn, and Juffer (2003).
Screening the entire data set for outliers revealed one study (Vasey, El
Hag, & Daleiden, 1996) that yielded an effect size that was larger than 3
standard deviations from the group means of each of the major effect size
categories used in this meta-analysis (within anxious group, within control
group, and between groups), and was thus removed from further analyses.
One additional study (Kyrios & Iob, 1998) yielded an effect size larger than
3 standard deviations from the mean effect size of the within-anxiety-group
data, and this particular effect size was also excluded from further analyses.
Therefore, the present meta-analysis was based on 172 samples reported in
142 journal articles. These 172 studies yielded 323 effects: 112 represent-
ing within-group attentional bias effects in anxious participants, 87 repre-
senting within-group attentional bias effects in control participants, and
124 representing between-groups differences in attentional bias between
anxious and control groups. A table describing the 172 studies included in
the meta-analyses may be obtained from us on request.
The results are organized into three main sections. In the first section, we
focus on the effects of procedural variables, namely, stimulus exposure
Table 1
Coding System for Individual Studies
Variable Coding description
N anxious Sample size for which a within-anxious-group effect is reported
N control Sample size for which a within-control-group effect is reported
Age group 0 children (under 18 years of age)
1 adults
Anxiety type 0 clinically diagnosed disorders
1 high self-reported anxiety
Type of anxiety disorder 0 generalized anxiety disorder
1 obsessive–compulsive disorder
2 panic disorder
3 post-traumatic stress disorder
4 social phobia
5 simple phobia
Type of anxiety measure
(self-reporting samples)
0 trait anxiety (State-Trait Anxiety Inventory)
1 state anxiety (State-Trait Anxiety Inventory)
2 other
Mood comorbidity 0 participants with comorbid mood disorder were excluded
1 participants with comorbid mood disorder were included
Experimental procedure
Paradigm 0 emotional Stroop
1 dot probe
2 modified Posner
Exposure time, general 0 subliminal
1 supraliminal
Exposure time, dot probe 0 subliminal
1 500 ms
2 1,000 ms
Stimulus type 0 words
1 pictures
Stroop design 0 blocked
1 random
time, experimental paradigm, and stimulus type, and the interactions
among them. In the second section, we describe effects of population-
related variables, namely, clinical versus nonclinical anxiety, age, type of
anxiety disorder, state versus trait anxiety, and mood disorder comorbidity,
and the interactions among them. In the third section, we report interactions
among the procedural and population-related moderators.
For each question, we assessed the magnitude of the combined within-
anxiety-group effect, the combined within-control-group effect, and the
difference between these two effects. In addition, we calculated the mag-
nitude of the difference in threat-related bias between the anxious and
control groups using the effect sizes of between-group comparisons. To-
gether, these analyses provide a comprehensive view of the state of affairs
regarding each of the pertinent questions. Table 2 provides combined
within-group effect sizes, CIs, p values, number of studies (k), and number
of participants (n) that were included in the analyses for the anxiety and
control groups. Table 3 provides the same data for the between-groups
analyses. For brevity reasons, only the essence of the findings concerning
each of the focal questions is reported in the text. A full report of all
statistics is provided in Tables 2 and 3.
Overall Effect of Threat-Related Bias
Across all the studies that reported within-group effects, the
combined effect size of the threat-related bias was significant in
anxious participants (k 112, n 2,263, d 0.45, p .01, CI
0.40, 0.49), and nonsignificant in nonanxious controls (k 87,
n 1,768, d 0.007, p .85, CI 0.06, 0.05). The difference
between the combined effect sizes of anxious and control partic-
ipants was significant (Q 87.56, p .01). Meta-analysis of the
studies that reported between-groups comparisons (k 125, n
2,963 and n 2,906 for anxious and control participants, respec-
tively) indicated an overall difference between the two groups
(d 0.41, p .01, CI 0.34, 0.48) that was comparable to the
within-group effect for anxious participants. Thus, a significant
threat-related bias was present in anxious participants but not in
nonanxious participants.
Procedural Moderators
Is Threat-Related Bias Stronger With Supraliminal
Exposure Times Than With Subliminal Exposure Times?
The combined effect sizes of threat-related bias in anxious
participants were significant both for studies using subliminal
exposure times (k 20, d 0.32, CI 0.20, 0.44) and for studies
using supraliminal exposure times (k 90, d 0.48, CI 0.42,
0.54). Although the combined effect size was slightly larger for
supraliminal than subliminal exposures, this difference was not
significant. Control participants showed no bias with either expo-
sure time, and the combined effects for anxious participants were
significantly larger than those for control participants with both
exposure times. Consistent with this finding, the between-groups
data revealed a significant difference between anxious and control
participants for both exposure times (ks 27 and 98, ds 0.37
and 0.42, CIs 0.22, 0.52, and 0.34, 0.50, for subliminal and for
supraliminal exposures, respectively).
Does the Magnitude of Threat-Related Bias Vary as a
Function of Experimental Paradigm (Emotional Stroop,
Dot Probe, Emotional Spatial Cuing)?
Anxious individuals displayed a significant threat-related bias
with each of the tested paradigms (k 70, d 0.49, CI 0.43,
0.56, with emotional Stroop; k 35, d 0.37, CI 0.28, 0.46,
with dot probe; and k 7, d 0.43, CI 0.23, 0.64, with
emotional spatial cuing). Control participants did not show a bias
with any of the paradigms. In addition, the combined within-
subject effect sizes of anxious participants were significantly
larger than those of control participants for each of the paradigms.
The between-groups comparisons revealed a similar pattern for
studies using the emotional Stroop (k 77,d 0.45, CI 0.36,
0.54) and those using the dot-probe (k 44,d 0.38, CI 0.26,
0.50) paradigm. By contrast, the combined between-groups effect
size of studies using the emotional spatial cuing paradigm was not
significant (k 4, d 0.009, CI 0.38, 0.40). It should be
noted, however, that the latter combined effect size was based on
four studies, two of which came from Broomfield and Turpin
(2005), who obtained negative validity effects (i.e., faster reaction
times on invalid-cue relative to valid-cue trials). In conclusion, the
dot-probe and emotional-Stroop paradigms were equally effective
in uncovering the between-groups bias, and evidence from the
emotional spatial cuing paradigm was equivocal.
Is the Threat-Related Bias Larger in Emotional Stroop
Studies Using Blocked Versus Random Designs?
Although both blocked- and random-design emotional Stroop
studies yielded significant effects of threat-related bias in anxious
participants, blocked presentations produced a significantly larger
combined effect size (k 31, d 0.69, CI 0.60, 0.78) than did
randomized presentations (k 30,d 0.35, CI 0.26, 0.44). Of
interest, a small but significant combined effect size was also
found for control participants in studies using a blocked design
(k 18, d 0.21, CI 0.10, 0.31). Notwithstanding, the
difference between anxious and nonanxious participants was sig-
nificant for both presentation modes, as reflected by the contrasts
between the within-group combined effects of anxious and control
participants and by the between-groups analyses.
Do Threat-Related Picture Stimuli Yield Larger Bias
Than Word Stimuli?
There was no difference between the combined effect sizes with
word stimuli and picture stimuli. Both types of stimuli produced a
significant threat-related bias in anxious participants (k 88,d
0.46, CI 0.40, 0.52, for words; k 24,d 0.43, CI 0.32,
0.54, for pictures) and no significant bias in nonanxious controls.
There was a significant difference between anxious and control
participants with both word stimuli and picture stimuli.
Experimental Paradigm Exposure Times
Does the magnitude of threat-related bias differ as a function of
exposure times in dot-probe experiments? In dot-probe experi-
ments, anxious participants showed a significant threat-related bias
with all exposure times (k 5,d 0.65, CI 0.42, 0.88, for
subliminal exposures; k 18, d 0.31, CI 0.20, 0.43, for
500-ms exposures; and k 7,d 0.29, CI 0.11, 0.47, for
exposures 1,000 ms), with no difference among the three
exposure conditions. However, subliminal exposures (k 5,d
0.65, CI 0.42, 0.88) yielded a significantly larger effect than
supraliminal exposures (500 ms and over, k 25, d 0.31, CI
0.21, 0.40, Q 4.12, p .05). Nonanxious control participants
did not show significant effects with supraliminal exposures (500
ms and 1,000 ms) but did show a significant bias away from
threat with subliminal exposures (k 4,d 0.28, CI 0.47,
0.09). In addition, the differences between anxious and control
participants were significant for subliminal and 500-ms exposures
but failed to reach significance with longer exposures ( 1,000
Does the magnitude of threat-related bias differ between sub-
liminal and supraliminal exposure times in emotional Stroop ex-
periments? Subliminal and supraliminal stimulus exposures re-
vealed significant combined effect sizes in anxious individuals and
nonsignificant combined effect sizes in controls. Here, the threat-
related bias in anxious participants was significantly larger with
supraliminal exposures (k 55,d 0.57, CI 0.50, 0.64) than
with subliminal exposures (k 15,d 0.23, CI 0.10, 0.35).
Comparison between the emotional Stroop, dot-probe, and emo-
tional spatial cuing paradigms with subliminal and supraliminal
exposure times. No emotional spatial cuing study used sublimi-
nal exposures. Contrasting combined effect sizes between emo-
tional Stroop studies and dot-probe studies with subliminal expo-
sures in anxious participants revealed a larger combined effect size
for dot-probe studies (Q 7.20, p .01). The same comparison
with supraliminal stimulus exposures revealed a larger effect size
for emotional Stroop studies than for dot-probe studies (Q 8.85,
p .01). Contrasts involving the emotional spatial cuing paradigm
were nonsignificant.
Stimulus Type Exposure Times
Do threat-related pictures yield larger effects than threat-
related words with supraliminal exposures? With supraliminal
exposures, both words (k 72,d 0.51, CI 0.44, 0.57) and
pictures (k 20,d 0.38, CI 0.26, 0.50) produced significant
bias effects in anxious, but not in control participants. There was
no difference in effect sizes between the two types of stimuli.
Do threat-related pictures yield larger effects than threat-
related words with subliminal exposures? The findings with
subliminal stimulus exposures were inconclusive. Analyses within
the anxious group showed that subliminally presented pictures
produced a significantly larger combined effect size (k 4, d
0.65, CI 0.42, 0.89) than did subliminally presented words (k
16, d 0.24, CI 0.15, 0.34). Surprisingly, the combined effect
size of the between-groups comparisons was significant for words
(k 21, d 0.42, CI 0.24, 0.61), but not for pictures (k 6,
d 0.19, CI 0.17, 0.54). Unfortunately, there were not enough
samples with subliminally presented pictures for control partici-
pants (k 3) to allow a complete analysis of this matter.
Experimental Paradigm Stimulus Type
Do threat-related pictures yield larger bias effects than threat-
related words in studies using the emotional Stroop paradigm?
Anxious participants showed a significant combined effect size
with words (k 66,d 0.51, CI 0.44, 0.58) and a nonsignif-
icant combined effect size with pictures (k 4,d 0.24, CI
0.02, 0.50). However, the difference between the two types of
stimuli was nonsignificant. In control participants, the bias was not
significant with words, and not enough emotional Stroop samples
used pictures (k 3).
The between-groups analyses revealed significant differences
between anxious and control participants in emotional Stroop
studies that used words (k 71, d 0.48, CI 0.41, 0.55) and
a nonsignificant combined effect size in the opposite direction for
studies that used pictures (k 5,d 0.25, CI 0.52, 0.01).
Words produced significantly larger interference effect than did
pictures (Q 14.44, p .01).
Do threat-related pictures yield larger bias effects than threat-
related words within studies using the dot-probe paradigm?
Threat-related bias in anxious participants was significant both
with words (k 20,d 0.35, CI 0.25, 0.46) and with pictures
(k 15, d 0.38, CI 0.26, 0.50), with no significant difference
between the two. Of importance, nonanxious control participants
displayed a significant bias away from threat in experiments using
words (k 17,d 0.14, CI 0.23, 0.05), and no bias with
Population-Related Moderators
Is Threat-Related Bias Stronger in Clinically Anxious
Participants Than in Participants With High Self-
Reported Anxiety?
Threat-related bias was significant in clinically diagnosed pop-
ulations (k 62, d 0.45, CI 0.38, 0.51) as well as in
populations with high self-reported anxiety (k 50,d 0.46,
CI 0.38, 0.54). The combined effect sizes of the two populations
did not differ.
Does the Magnitude of Threat-Related Bias Differ as a
Function of Anxiety Disorder?
We compared the combined effect sizes of groups of studies that
focused on specific anxiety disorders using threat stimuli that were
congruent with the concerns of the specific disorder studied. In-
cluded were studies of GAD (k 11), OCD (k 6), panic
disorder (k 7), PTSD (k 22), social phobia (k 8), and simple
phobia (k 5). Tables 2 and 3 provide the detailed statistics of
these comparisons. The results indicate that threat-related bias was
significant for all the clinical disorders tested, with effect sizes
ranging from 0.36 to 0.59. The combined effect sizes did not differ
among the various disorders.
Does Comorbidity Between Anxiety and Mood Disorders
Affect the Magnitude of Threat-Related Bias in Clinically
Anxious Populations?
The combined effect sizes of bias were significant both for
studies that excluded participants with comorbid mood disorders
(k 17, d 0.43, CI 0.33, 0.53) and for studies that did not
Table 2
Meta-Analytic Results of Within-Group Threat-Related Biases for Anxious Participants (k 112 Outcomes) and Control Participants (k 87 Outcomes)
Within anxious Within control Anxious vs. control
kN d 85% CI Q
pkN d 85% CI Qp Q
Total data set 112 2.263 0.45
0.40, 0.49 87 1,768 0.007 0.06, 0.05 87.56 .0001
Procedural moderators
Stimulus exposure time
Supraliminal 92 1,837 0.48
0.42, 0.54 3.08 .08 72 1,427 0.009 0.04, 0.06 1.41 .23 75.03 .0001
Subliminal 20 426 0.32
0.20, 0.44 15 641 0.09 0.20, 0.02 14.48 .0001
Stroop 70 1,467 0.49
0.43, 0.56 2.39 .30 47 973 0.06 0.001, 0.12 5.92 .05 40.25 .0001
Dot probe 35 659 0.37
0.28, 0.46 33 655 0.09 0.16, 0.01 38.96 .0001
Posner 7 137 0.43
0.23, 0.64 7 140 0.11 0.27, 0.05 7.32 .01
Stroop design
Blocked 31 645 0.69
0.60, 0.78 14.78 .0001 18 356 0.21
0.10, 0.31 5.94 .05 19.23 .0001
Random 30 588 0.35
0.26, 0.44 22 432 0.03 0.12, 0.06 21.94 .0001
Stimulus type
Words 88 1,798 0.46
0.40, 0.52 0.14 .71 63 1,304 0.006 0.06, 0.05 0.02 .90 64.06 .0001
Pictures 24 465 0.43
0.32, 0.54 24 464 0.02 0.10, 0.07 21.77 .0001
Paradigm Exposure
Dot probe—subliminal 5 93 0.65
0.42, 0.88 4.04 .13 4 79 0.28
0.47, 0.09 3.37 .19 20.13 .0001
Dot probe—500 ms 18 339 0.31
0.20, 0.43 19 375 0.06 0.15, 0.02 13.86 .0001
Dot probe—1,000 ms 7 141 0.29
0.11, 0.47 6 117 0.03 0.13, 0.18 2.72 .10
Stroop—subliminal 15 333 0.23
0.10, 0.35 11.57 .001 11 626 0.02 0.15, 0.11 1.09 .30 4.95 .05
Stroop—supraliminal 55 1,134 0.57
0.50, 0.64 36 711 0.09 0.01, 0.17 32.43 .0001
Stimulus Type Exposure
Supraliminal—words 72 1,433 0.51
0.44, 0.57 1.73 .19 51 1,018 0.01 0.05, 0.07 0.04 .85 57.04 .0001
Supraliminal—pictures 20 404 0.38
0.26, 0.50 21 409 0.001 0.09, 0.09 16.14 .0001
Subliminal—words 16 365 0.24
0.15, 0.34 5.25 .05 12 286 0.08 0.21, 0.06 10.68 .001
Subliminal—pictures 4 61 0.65
0.42, 0.89
Stimulus Type Paradigm
Stroop—words 66 1,372 0.51
0.44, 0.58 2.12 .15 44 901 0.06 0.008, 0.13 36.71 .0001
Stroop—pictures 4 95 0.24 0.02, 0.50
Dot probe—words 20 388 0.35
0.25, 0.46 0.05 .82 17 363 0.14
0.23, 0.05 1.59 .21 35.90 .0001
Dot probe—pictures 15 271 0.38
0.26, 0.50 16 292 0.03 0.12, 0.07 10.99 .0001
Population-related moderators
Anxiety type
Clinical anxiety 62 1,326 0.45
0.38, 0.51 0.03 .86 40 872 0.05 0.02, 0.11 2.59 .11 35.05 .0001
High self-reported anxiety 50 937 0.46
0.38, 0.54 47 896 0.06 0.12, 0.006 50.27 .0001
Type of anxiety disorder
GAD 11 202 0.56
0.37, 0.66 4.57 .47
OCD 6 76 0.45
0.25, 0.65
Panic disorder 7 170 0.50
0.34, 0.66
PTSD 22 502 0.36
0.27, 0.46
Social phobia 8 200 0.59
0.43, 0.74
Simple phobia 5 128 0.36
0.43, 0.74
Comorbid mood disorders
Excluded 17 379 0.43
0.33, 0.53 0.17 .68
Included 24 433 0.47
0.38, 0.56
State vs. trait anxiety
State 5 92 0.65
0.42, 0.88 2.48 .12 5 103 0.19 0.33, 0.05 1.23 .27 30.45 .0001
Trait 29 517 0.38
0.28, 0.48 30 545 0.07 0.13, 0.008 30.82 .0001
Table 2 (continued)
Within anxious Within control Anxious vs. control
kN d 85% CI Q
pkN d 85% CI Qp Q
Population-related moderators (continued)
Adults 101 2,013 0.45
0.39, 0.50 0.24 .63 78 1,580 0.009 0.06, 0.04 0.00 .99 80.16 .0001
Children 11 250 0.50
0.34, 0.66 9 188 0.008 0.15, 0.13 7.35 .01
Anxiety Type Age
Adults—clinical 54 1,141 0.45
0.38, 0.52 0.04 .84 34 751 0.05 0.02, 0.13 2.35 .13 29.42 .0001
Adults—nonclinical 47 872 0.43
0.36, 0.51 44 829 0.06 0.12, 0.01 48.84 .0001
Children—clinical 8 185 0.40
0.24, 0.56 6 121 0.02 0.16, 0.21 4.99 .05
Procedure Population
Exposure Anxiety Type
Clinical—supraliminal 48 1,013 0.51
0.44, 0.57 10.10 .001 30 630 0.08 0.007, 0.17 1.01 .32 32.09 .0001
Clinical—subliminal 14 313 0.22
0.11, 0.33 10 242 0.04 0.19, 0.11 5.62 .05
Nonclinical—supraliminal 44 824 0.45
0.35, 0.54 0.67 .41 42 797 0.04 0.10, 0.01 1.20 .27 38.79 .0001
Nonclinical—subliminal 6 131 0.61
0.34, 0.87 5 99 0.18 0.35, 0.01 12.51 .0001
Paradigm Anxiety Type
Clinical—Stroop 45 1,006 0.48
0.41, 0.55 2.20 .13 26 580 0.09 0.001, 0.19 .65 .42 22.56 .0001
Clinical—dot probe 16 302 0.34
0.22, 0.46 13 272 0.002 0.13, 0.13 9.31 .01
Nonclinical—Stroop 25 461 0.51
0.38, 0.64 0.71 .70 21 393 0.02 0.06, 0.10 4.26 .12 16.44 .0001
Nonclinical—dot probe 19 357 0.40
0.25, 0.55 20 383 0.14
0.22, 0.06 30.68 .0001
Nonclinical—Posner 6 119 0.48
0.23, 0.74 6 120 0.06 0.20, 0.09 5.78 .05
Stimulus Type Anxiety Type
Clinical—words 57 1,221 0.46
0.40, 0.52 2.60 .11 36 793 0.03 0.05, 0.11 1.69 .19 39.95 .0001
Clinical—pictures 5 105 0.22 0.02, 0.42 4 79 0.26 0.02, 0.50 .02 .88
Nonclinical—words 31 577 0.45
0.34, 0.63 0.09 .77 27 511 0.05 0.12, 0.02 0.05 .82 22.57 .0001
Nonclinical—pictures 19 360 0.49
0.34, 0.57 20 385 0.07 0.15, 0.02 30.63 .0001
Paradigm Age
Adults—Stroop 64 1,330 0.47
0.40, 0.53 1.20 .55 42 864 0.08 0.008, 0.14 7.09 .05 33.00 .0001
Adults—dot probe 30 546 0.38
0.29, 0.48 29 576 0.10 0.18, 0.02 39.96 .0001
Adults—Posner 7 137 0.43
0.24, 0.63 7 140 0.11 0.27, 0.05 7.32 .01
Children—Stroop 6 137 0.69
0.40, 0.99 1.53 .22 5 109 0.06 0.20, 0.07 0.28 .59 7.16 .01
Children—dot probe 5 113 0.32 0.009, 0.64 4 79 0.02 0.16, 0.20 1.15 .28
Age Exposure
Supraliminal—adults 81 1,587 0.48
0.42, 0.54 0.05 .83 63 1,239 0.01 0.04, 0.07 0.05 .83 67.77 .0001
Supraliminal—children 11 250 0.50
0.34, 0.67 9 188 0.01 0.14, 0.12 7.35 .01
Subliminal—adults 20 426 0.32
0.20, 0.44 15 641 0.09 0.20, 0.02 14.48 .0001
Stimulus Type Age
Words—adult 79 1,605 0.43
0.37, 0.49 3.78 .05 56 1,163 0.002 0.06, 0.06 0.05 .83 50.64 .0001
Words—children 9 193 0.68
0.51, 0.85 7 141 0.03 0.19, 0.14 18.00 .0001
Pictures—adults 17 408 0.51
0.39, 0.63 17 417 0.035 0.13, 0.06 26.25 .0001
Note. Dashes indicate k 4 (subcategories with k 4 studies were not tested).
Q for comparison between subcategories of a moderator.
Q for comparison between anxious and nonanxious control.
p .05.
p .01.
p .001.
p .0001.
Table 3
Meta-Analytic Results of Between-Groups Comparisons (k 125 Outcomes)
Moderator k
nd 85% CI Q
Total data set 125 2,963 2,906 0.41
0.34, 0.48
Procedural moderators
Stimulus exposure time
Supraliminal 98 2,337 2,295 0.42
0.34, 0.50 0.20 .65
Subliminal 27 593 583 0.37
0.22, 0.52
Stroop 77 1,988 1,936 0.45
0.36, 0.54 2.71 .26
Dot probe 44 889 887 0.38
0.26, 0.50
Posner 4 86 83 0.009 0.38, 0.40
Stroop design
Blocked 30 784 716 0.56
0.42, 0.70 1.84 .18
Random 39 1,015 1,027 0.38
0.27, 0.50
Stimulus type
Words 98 2,438 2,361 0.44
0.36, 0.52 1.66 .20
Pictures 26 476 494 0.28
0.13, 0.44
Paradigm Exposure
Dot probe—subliminal 6 126 123 0.56
0.17, 0.95 1.12 .57
Dot probe—500 ms 25 469 472 0.41
0.22, 0.60
Dot probe—1,000 ms 11 210 189 0.22 0.08, 0.51
Stroop—subliminal 21 467 460 0.32
0.17, 0.48 1.83 .18
Stroop—supraliminal 56 1,521 1,476 0.50
0.40, 0.59
Stimulus Type Exposure
Supraliminal—words 77 1,929 1,868 0.44
0.35, 0.52 0.84 .36
Supraliminal—pictures 20 392 404 0.32
0.16, 0.49
Subliminal—words 21 509 493 0.42
0.24, 0.61 0.74 .39
Subliminal—pictures 6 84 90 0.19 0.17, 0.54
Stimulus Type Paradigm
Stroop—words 71 1,870 1,814 0.48
0.41, 0.55 14.44 .0001
Stroop—pictures 5 87 89 0.25 0.52, 0.01
Dot probe—words 24 502 483 0.37
0.20, 0.53 0.04 .85
Dot probe—pictures 20 396 386 0.40
0.22, 0.58
Population-related moderators
Anxiety type
Clinical anxiety 50 1,229 1,138 0.50
0.38, 0.61 1.94 .16
High self-reported anxiety 75 1,716 1,750 0.36
0.27, 0.45
Type of anxiety disorder
Generalized anxiety disorder 12 221 236 0.55
0.37, 0.73
Obsessive-compulsive disorder 71 57
Panic disorder 8 188 174 0.52
0.31, 0.73
Post-traumatic stress disorder 11 278 216 0.46
0.28, 0.64
Social phobia 5 153 169 0.46
0.22, 0.71
Simple phobia 8 233 192 0.53
0.33, 0.73
State vs. trait anxiety
Trait 36 640 648 0.36
0.22, 0.51 0.02 .88
State 8 140 141 0.33 0.04, 0.62
Comorbid mood disorders
Excluded 14 319 332 0.45
0.29, 0.60 0.89 .35
Included 17 368 340 0.58
0.44, 0.73
Adults 108 2,298 2,257 0.42
0.34, 0.49 0.09 .77
Children 17 647 631 0.38
0.19, 0.56
Anxiety Type Age
Adults—clinical 42 1,039 1,002 0.49
0.37, 0.61 1.13 .29
Adults—nonclinical 66 1,259 1,255 0.37
0.27, 0.47
Children—clinical 8 190 196 0.52
0.30, 0.74 2.16 .14
Children—nonclinical 9 457 495 0.23 0.04, 0.41
Procedure Population
Exposure Anxiety Type
Clinical—supraliminal 37 899 815 0.57
0.48, 0.66 5.43 .05
Clinical—subliminal 13 330 323 0.29
0.14, 0.44
exclude participants with comorbid mood disorders (k 24, d
0.47, CI 0.38, 0.56). There was no difference between these two
combined effect sizes, suggesting that co-occurrence of mood
disorders with anxiety does not play a major role in the threat-
related bias of anxious individuals.
Is There a Difference in the Threat-Related Bias When
the High- and Low-Anxiety Groups Are Based on State
Versus Trait Anxiety?
In these analyses, only studies that relied on Spielberger’s
state–trait anxiety scales (Spielberger, Gorsuch, Lushene, Vagg, &
Jacobs, 1983) were included. For anxious participants, studies that
relied on state anxiety for assigning participants to anxious versus
nonanxious groups produced a somewhat larger combined within-
group effect size (k 5,d 0.65, CI 0.42, 0.88) than did
studies that relied on trait anxiety (k 29, d 0.38, CI 0.28,
0.48). However, the difference between these combined effect
sizes was not significant.
The results of the between-groups analyses revealed a somewhat
different picture. A significant difference between anxious and
control participants was found in studies that relied on trait anxiety
(k 36, d 0.36, CI 0.22, 0.51), but not in studies that relied
on state anxiety (k 8, d 0.33, p .10, CI 0.04, 0.62). Yet,
the difference between the combined effect sizes was again not
Do Children and Adults Show a Similar Threat-Related
Threat-related bias was significant in anxious adults (k 101.
d 0.45, CI 0.39, 0.50) and in anxious children (k 11, d
0.50, CI 0.34, 0.66) and did not differ between the two groups.
Clinical Diagnosis Versus High Self-Reported Anxiety by
The bias in anxious adults was significant both for clinical
populations (k 54,d 0.45, CI 0.38, 0.52) and for popula-
tions with high self-reported anxiety (k 47, d 0.43, CI 0.36,
0.51) and did not differ between the two population types. There
were not enough samples to conduct within-group meta-analyses
with children (k 3 for anxious children and k 3 for nonanxious
control children in studies relying on self-reported anxiety).
Table 3 (continued)
Moderator k
nd 85% CI Q
Procedure Population (continued)
Exposure Anxiety Type
Nonclinical—supraliminal 61 1,453 1,490 0.34
0.24, 0.45 0.31 .58
Nonclinical—subliminal 14 293 260 0.44
0.22, 0.66
Paradigm Anxiety Type
Clinical—Stroop 33 892 797 0.54
0.44, 0.64 1.54 .46
Clinical—dot probe 17 337 341 0.40
0.25, 0.54
Nonclinical—Stroop 44 1,096 1,139 0.38
0.26, 0.51 0.18 .60
Nonclinical—dot probe 27 534 528 0.37
0.21, 0.53
Nonclinical—Posner 4 86 83 0.009 0.41, 0.43
Stimulus Type Anxiety Type
Clinical—words 43 1,090 977 0.53
0.45, 0.61 6.57 .01
Clinical—pictures 6 108 128 0.09 0.15, 0.32
Nonclinical—words 55 1,348 1,384 0.37
0.25, 0.48 0.02 .89
Nonclinical—pictures 20 368 366 0.34
0.16, 0.53
Paradigm Age
Adults—Stroop 66 1,478 1,408 0.48
0.38, 0.57 3.40 .18
Adults—dot probe 38 734 766 0.36
0.23, 0.49
Adults—Posner 4 86 83 0.009 0.39, 0.41
Children—Stroop 11 510 528 0.27
0.10, 0.44 1.39 .24
Children—dot probe 6 137 103 0.53
0.26, 0.79
Age Exposure
Supraliminal—adults 81 1,705 1,674 0.43
0.35, 0.52 0.21 .64
Supraliminal—children 17 647 631 0.37
0.19, 0.55
Subliminal—adults 27 593 583 0.37
0.22, 0.52
Stimulus Type Age
Words—adult 83 1,848 1,777 0.44
0.36, 0.53 0.10 .76
Words—children 15 590 584 0.40
0.21, 0.59
Pictures—adults 24 419 447 0.29
0.10, 0.49
Note. Dashes indicate k 4 (subcategories with k 4 studies were not tested).
Q for comparison between subcategories of moderator.
p .05.
p .01.
p .001.
p .0001.
The between-groups data revealed a significant difference be-
tween anxious and control adults in both the clinically diagnosed
samples (k 42, d 0.49, CI 0.37, 0.61) and the samples
relying on self-reported anxiety (k 66, d 0.37, CI 0.27,
0.47). In children, a significant difference between the anxious and
control groups was found in the clinically diagnosed samples (k
8, d 0.52, CI 0.30, 0.74), but not in the samples relying on
self-reported anxiety (k 9, d 0.23, p .07, CI 0.04, 0.41).
However, the between-groups effect sizes did not differ between
the two types of child population.
Interactions Between Procedural and Population-Related
Does Stimulus Exposure Time Differentially Modulate the
Bias in Clinically Diagnosed Participants Versus in
Participants With High Self-Reported Anxiety?
In clinically anxious populations, threat-related bias was signif-
icant both with supraliminal exposures (k 48,d 0.51, CI
0.44, 0.57) and with subliminal exposures (k 14, d 0.22, CI
0.11, 0.33). The combined effect size was larger for supraliminal
exposures than for subliminal exposures (Q 10.10, p .01).
Consistent with this finding, the between-groups analyses showed
significant differences between clinically anxious and control in-
dividuals both with supraliminal exposures (k 37, d 0.57,
CI 0.48, 0.66) and with subliminal exposures (k 13, d 0.29,
CI 0.14, 0.44) and a larger effect size for the supraliminal than
for the subliminal exposures (Q 5.43, p .05).
In studies relying on self-reported anxiety, threat-related bias
was also significant both for supraliminal exposures (k 44, d
0.45, CI 0.35, 0.54) and for subliminal exposures (k 6, d
0.61, CI 0.34, 0.87), but the difference between the two expo-
sures was not significant.
A direct comparison of the combined effect sizes for clinically
anxious individuals and individuals with high self-reported anxiety
revealed that the effect sizes of the two groups did not differ with
supraliminal exposures, but that with subliminal stimulus expo-
sures the combined effect size was significantly smaller for clin-
ically anxious participants than for participants with high self-
reported anxiety (Q 5.49, p .05).
Does Experimental Paradigm Differentially Modulate
Threat-Related Bias in Clinically Diagnosed Participants
Versus in Participants With High Self-Reported Anxiety?
With clinically diagnosed participants, the bias was significant
and did not differ between emotional Stroop studies (k 45, d
0.48, CI 0.41, 0.55) and dot-probe studies (k 16, d 0.34,
CI 0.22, 0.46). There was only one study using the emotional
spatial cuing paradigm with clinically diagnosed participants
(Amir, Elias, Klumpp, & Przeworski, 2003), which precluded
further analysis with this paradigm. In control participants for
clinical studies, there was no threat-related bias.
In participants with high self-reported anxiety, the combined
effect sizes of the bias were significant and did not differ for
emotional Stroop studies (k 25, d 0.51, CI 0.38, 0.64),
dot-probe studies (k 19, d 0.40, CI 0.25, 0.55), or emo-
tional spatial cuing studies (k 6, d 0.48, CI 0.23, 0.74). In
control participants, there was no significant threat-related bias in
either emotional Stroop studies or emotional spatial cuing studies,
whereas a significant bias away from threat was found in dot-probe
studies (k 20, d 0.14, CI 0.22, 0.06). However, none
of the between-paradigms differences were significant.
Do Words and Pictures Differentially Modulate Threat-
Related Bias in Clinical Participants Versus in
Participants With High Self-Reported Anxiety?
In clinically diagnosed participants, the bias was significant with
words (k 57, d 0.46, CI 0.40, 0.52), but not with pictures.
In addition, the difference between the combined effect sizes from
the within-group analyses in clinically anxious versus control
participants was significant for studies using words (Q 39.95,
p .01), but not for studies using pictures. In line with these
findings, the between-groups effect was also significant with
words (k 43, d 0.53, CI 0.45, 0.61) and nonsignificant with
pictures, with a significant difference between the two effects
(Q 6.57, p .01).
In studies relying on self-reported anxiety, the threat-related bias
was significant both with words (k 31, d 0.45, CI 0.34,
0.63) and with pictures (k 19, d 0.49, CI 0.34, 0.57), with
no difference between the two effects.
Does the Type of Experimental Paradigm Differentially
Modulate the Bias in Children Versus in Adults?
The pattern of results regarding the effects of experimental
paradigm in adults was similar to that of the full data set. Threat-
related bias in anxious adults was significant in emotional Stroop
(k 64, d 0.47, CI 0.40, 0.53), dot-probe (k 30, d 0.38,
CI 0.29, 0.48), and emotional spatial cuing studies (k 7, d
0.43, CI 0.24, 0.63), with no difference between the paradigms.
In control adults, there was no threat-related bias for any of the
experimental paradigms. However, control participants in dot-
probe and emotional spatial cuing studies showed a nonsignificant
bias away from threat, whereas control participants in emotional
Stroop studies showed a nonsignificant slowing with threat relative
to neutral stimuli. This resulted in a significant difference among
the combined effect sizes of the three paradigms (Q 5.91,
p .05).
A different pattern of results emerged for children. The threat-
related bias was significant in emotional Stroop (k 6, d 0.69,
CI 0.40, 0.99), but not in dot-probe studies (k 5, d 0.32,
CI 0.009, 0.64). The within-group combined effect size was
significantly larger in anxious than in control participants in emo-
tional Stroop studies (Q 7.16, p .01), but there was no
difference between the two child populations in dot-probe studies.
No emotional spatial cuing studies were conducted with children.
Unlike the results of the within-group comparisons, the between-
groups data showed a significant difference between anxious and
control children both in emotional Stroop (k 11, d 0.27, CI
0.10, 0.44) and in dot-probe studies (k 6, d 0.53, CI 0.26,
Do Stimulus Exposure Time and Stimulus Type
Differentially Modulate the Bias in Children and Adults?
There were no studies of children using subliminal exposures.
For supraliminal exposures, the combined effect sizes of anxious
adults (k 81, d 0.48, CI 0.42, 0.54), and anxious children
(k 11, d 0.50, CI 0.34, 0.67) were significant and did not
differ from each other.
Only two studies with children used picture stimuli, thus pre-
cluding a comparison between adults and children on this variable.
For word stimuli, the bias was significant both for anxious adults
(k 79, d 0.43, CI 0.37, 0.49) and for anxious children (k
9, d 0.68, CI 0.51, 0.85), with children showing a larger effect
size than adults (Q 3.78, p .05).
The main conclusion of this set of meta-analytic studies is that
the threat-related bias is a robust phenomenon in anxious individ-
uals and does not exist in nonanxious individuals. Although the
threat-related bias in anxious individuals holds under a variety of
experimental conditions, and in different types of anxious popu-
lations, this consistent phenomenon is of low-to-medium effect
size. The meta-analytic finding for the anxious participants cannot
be reduced to insignificance in the next 11,339 studies, even if
those studies yielded only null results. This number is 20 times as
large as Rosenthal’s (1991) fail-safe number, 5 k 10 570 (k
number of studies included), such that the file-drawer problem is
not of concern here. In fact, this large fail-safe number points to
the diminishing returns to be expected from further studies that
only focus on establishing the presence of a threat-related bias in
anxious groups. New directions for further research in this exciting
area are needed based on the premise of a moderate attentional bias
in anxious participants.
No Threat-Related Bias in Nonanxious Individuals
The absence of a threat-related bias in nonanxious participants
suggests that threat-related material presented in controlled exper-
imental environments does not summon the attention of nonanx-
ious individuals more than does neutral material. This finding
appears to be inconsistent with previous literature suggesting that
all humans are prewired to automatically orient toward potential
threat in the environment (e.g., LeDoux, 1995; Ohman, 1993) or
with visual search studies with normative samples showing that a
face displaying a threatening emotion is detected faster than a face
displaying either a neutral or a happy expression (e.g., Eastwood,
Smilek, & Merikle, 2001; Fox, Lester, Russo, Bowles, & Dutton,
2000; Hansen & Hansen, 1988).
Two possible explanations for this discrepancy might be sug-
gested but should be rejected on the basis of the present meta-
analysis. First, many of the studies on nonclinical populations
reported in the present meta-analysis used low-anxious partici-
pants (e.g., bottom quartile of the distribution of a normative
population on an anxiety index) as opposed to groups from the
general population with an average anxiety level that were tested
in studies investigating attention to threat with no reference to
anxiety. Thus, one could argue that the control groups tested in the
present analyses represent a group of people with unusually low
levels of anxiety. This explanation is unlikely, however, because
null bias effects were found also for control participants in studies
of clinical populations, where such selection bias did not prevail.
Second, it could be argued that the bias toward threat stimuli in the
general population may be specific to naturalistic or biologically
valid stimuli and not hold for word stimuli. However, the present
meta-analysis shows that naturalistic stimuli (threat-related pic-
tures) produced null effects in the control group just as did threat-
related word stimuli.
It is noteworthy that a significant threat-related bias in nonanx-
ious participants did emerge in one experimental condition,
namely, in the blocked-design emotional Stroop. It may be the case
that a threat-related bias may be detected in nonanxious partici-
pants with experimental designs that involve cumulative exposure
to threat-related stimuli and thereby produce stronger perceived
threat. Consistent with this interpretation, it has been suggested
that whereas anxious individuals show an attentional bias even
with mildly threatening stimuli, nonanxious individuals show a
bias only with high levels of threat (e.g., Mogg & Bradley, 1999b;
Wilson & MacLeod, 2003). Unfortunately, this hypothesis could
not be tested in the present meta-analysis because a comparison of
stimuli’s threat levels across studies was impossible.
In addition, the results partially support the suggestion that
under mild threat conditions, nonanxious individuals in fact show
avoidance of threat-related stimuli and shift attention away from
them (e.g., MacLeod et al., 1986; Williams et al., 1988). A sig-
nificant bias away from threat was observed in control individuals
under specific conditions in dot-probe studies (i.e., with subliminal
exposure times, with word stimuli, and in nonclinical populations).
However, these findings should be interpreted with caution, be-
cause the effect sizes are small and not consistent across studies.
Stimulus Awareness Modulates the Bias in Opposite
Directions in Emotional Stroop and Dot-Probe Studies
The meta-analysis indicates that subliminally perceived threat
summons anxious individuals’ attention, which entails that the
attentional bias in anxiety is not contingent on conscious percep-
tion of threat. However, the combined effect size of the bias was
smaller with subliminal exposures than with supraliminal expo-
sures, which indicates that part of the threat-related bias in anxious
individuals does result from processes that require conscious per-
Furthermore, preconscious and conscious processes were found
to be differentially tapped by the emotional Stroop and the dot-
probe paradigms. In emotional Stroop studies, although combined
effect sizes in anxious participants were significant both with
subliminal and with supraliminal exposures, the latter was signif-
icantly larger. This finding suggests that conscious processes play
a prominent role in this paradigm. The reverse pattern was ob-
served in dot-probe studies. Subliminal exposures yielded a com-
bined effect size in anxious individuals that was almost twice as
large as that yielded by supraliminal exposures, suggesting that
conscious processes contribute relatively little to the threat-related
attentional bias reported in dot-probe studies. This pattern of
results is consistent with the claim that the bias observed using the
emotional Stroop reflects relatively late, controlled processes,
whereas the bias revealed using the dot-probe paradigm reflects
earlier attentional processes (see MacLeod et al., 1986).
Naturalistic Threat Stimuli Produce Larger Bias Than
Word Stimuli Only When Presented Subliminally
Overall, the findings do not support the claim that naturalistic
threat stimuli (e.g., pictures of angry or fearful faces, spiders, etc.)
are more potent than word stimuli for the assessment of a threat-
related bias in anxious individuals. Indeed, the combined effect
sizes of studies using picture stimuli versus words did not differ.
With subliminal exposures, however, naturalistic stimuli did show
the expected superiority, as the combined effect size was almost 3
times as large with naturalistic stimuli as with word stimuli. This
result may reflect the fact that relative to sensory processing,
semantic processing requires longer exposures. Thus, when
enough time is provided for processing of threat-related stimuli,
both words and naturalistic stimuli induce similar bias effects, but
when extremely fast, automatic processing is required, biologically
salient stimuli are more potent in eliciting a threat-related bias.
This interpretation is consistent with descriptions of a fast and
direct neural relay of sensory information to the mammalian amyg-
dala, which is dedicated to the processing of potentially dangerous
events in the natural environment (LeDoux, 1995; LeDoux, Cic-
chetti, Xagoraris, & Romanski, 1990).
However, two additional observations call for cautious consid-
eration of this conclusion. First, despite the inherent problems
associated with the interpretation of between-subjects comparisons
(as explained in the introduction), these revealed a significant
combined effect for subliminally presented words and a nonsig-
nificant combined effect for subliminally presented pictures. Sec-
ond, further exploration of the data from subliminal exposures in
within-subject analyses suggests a possible confound between
stimulus type and paradigm. There were more studies using the
dot-probe paradigm than studies using the emotional Stroop par-
adigm with naturalistic stimuli (3 vs. 1), whereas there were more
studies using the Stroop paradigm than studies using the dot-probe
paradigm with words (14 vs. 2). Because the dot-probe paradigm
tends to produce larger subliminal effects than does the Stroop
paradigm, the paradigm used rather than the type of stimulus might
account for the observed advantage of naturalistic stimuli over
word stimuli with subliminal exposures. Because of the small
number of studies currently available in each relevant cell, con-
clusions that are more definitive must await further investigation.
The Threat-Related Bias Is Similar in Clinically Anxious
and Nonclinical High-Anxious Participants
The meta-analysis did not support the hypothesis that clinically
anxious individuals present a more robust bias than do individuals
with high levels of self-reported anxiety. Indeed, we found equiv-
alent combined effect sizes of threat-related bias in the two pop-
ulations. These findings suggest that an official clinical cutoff is of
little significance with regard to biased attentional processes in
anxious individuals, and that milder forms of anxiety are sufficient
for triggering the full potential of the bias. Thus, it appears that the
existence of a threat-related bias may not suffice to determine
whether a highly anxious individual will develop an anxiety dis-
order. To determine whether the bias is nonetheless an etiological
factor in clinical anxiety, it will be useful to obtain individual data
in addition to group means in order to estimate what proportion of
clinically anxious participants does not present the bias at all.
In addition, there is some indication in the data that the threat-
related bias in clinical anxiety might be dependent on conscious
processing of the threat-related stimuli. Indeed, clinically anxious
participants showed larger combined effects size with supraliminal
exposures than with subliminal exposures, whereas nonclinically
anxious participants showed equivalent combined effect sizes for
the two exposure conditions. This finding may have important
practical implications for future interventions aiming at a system-
atic reduction of the bias in clinically anxious patients, suggesting
that focusing on the conscious aspects of the bias may be most
beneficial. However, this conclusion should be approached with
caution because scrutiny of the data showed that with clinical
samples the Stroop paradigm, which yields larger bias effects with
supraliminal stimuli, was more frequently used than the dot-probe
paradigm, which yields larger bias effects with subliminal stimuli.
With nonclinical samples, the imbalance in the relative number of
studies in each paradigm was much smaller.
The Magnitude of the Threat-Related Bias Is Similar in
All Anxiety Disorders
Considering the diverse phenotypes of the different anxiety
disorders, the finding of a similar-size bias in all the anxiety
disorders studied is striking. This finding might indicate that the
bias is related to a core anxiety component that is common to all
anxiety disorders as well as to nonclinical anxiety. Although this
idea cannot accommodate the often reported finding that no atten-
tional bias is found in individuals with depression despite high
comorbidity with clinical anxiety and high levels of reported
anxiety (for a review, see Mogg & Bradley, 2005), it is noteworthy
that the present meta-analysis showed that whether or not partic-
ipants with depression were included in the anxious group did not
modulate the attentional bias effect.
Children and Adults Show a Similar Pattern of Threat-
Related Bias
The meta-analysis shows that the combined effect sizes in
anxious children and in anxious adults were both significant and
did not differ from each other. The results further show that
nonanxious children and nonanxious adults show no threat-related
bias. However, because there were not enough studies with chil-
dren to allow a more sensitive breakdown of the data by age group
and because many of the studies with children used a wide age
range, a more detailed description of the developmental course of
attentional bias in children must await further research.
Cognitive Mechanisms Underlying Threat-Related Bias in
As reviewed in the introduction, several theoretical views of the
mechanisms underlying threat-related bias in anxiety have been
proposed. Although a common aspect of these views is an empha-
sis on the role of preattentive and attention-allocation processes in
trait anxiety, there is no consensus as to the exact mechanisms
underlying these biases.
Williams at al. (1988, 1997) have proposed that two cognitive
mechanisms are responsible for the threat-related bias in anxious
individuals: an affective decision mechanism (ADM) and a re-
source allocation mechanism (RAM; or task demand unit in the
1997 model). The function of the ADM is to assess the threat value
of stimuli. The RAM receives input from the ADM and determines
resource allocation. According to Williams et al., individual dif-
ferences in the RAM underlie individual differences in trait anx-
iety, with high-trait-anxious individuals showing a permanent ten-
dency to orient toward threat and low-trait-anxious individuals
tending to shift attention away from threat.
Mogg and Bradley (1998) proposed a cognitive-motivational
model in which individual differences in trait anxiety concern the
reactivity of a valence evaluation system (VES) that is similar to
Williams et al.’s (1988, 1997) ADM. According to Mogg and
Bradley (1998), the VES is more sensitive in high-trait-anxious
individuals, that is, stimuli that nonanxious individuals tag as
nonthreatening are tagged as threatening by anxious individuals.
Output from the VES feeds into a goal engagement system that
determines the allocation of resources for cognitive processing and
action. In this model, if a high threat value is assigned to a
stimulus, interruption of ongoing activity is automatically deter-
mined in both high- and low-anxious individuals. Other research-
ers (e.g., Wells & Matthews, 1994) have advanced a completely
different view, focusing on voluntary, strategic processes in me-
diating biases in anxiety.
Although the findings from the present meta-analysis do not
offer clear-cut support for one model over the other, they undoubt-
edly challenge some of the outcomes predicted by each of them.
First, our findings suggest that Williams et al.’s (1988, 1997) claim
that low-trait-anxious individuals show bias away from threat is at
best a very weak phenomenon. Second, the meta-analytic data
show that individual differences in anxiety are most probably
driven both by preattentive threat detection biases, as reflected in
the unequivocal evidence for a bias with stimuli outside awareness,
and by later resource allocation mechanisms and top-down pro-
cesses, as reflected in the larger effect size for consciously per-
ceived relative to subliminally exposed threat-related stimuli. The
distinction between separate contributions of unconscious and
conscious processes to threat processing in anxiety is further
validated by the fact that stimulus awareness modulates the bias in
opposite directions in emotional Stroop versus dot-probe studies.
Thus, although the present findings provide some support for
Mogg and Bradley’s (1998), Williams et al.’s (1988, 1997), and
Wells and Matthews’s (1994) models, they also suggest that strong
claims that bias in only one stage of processing accounts for the
attentional bias in anxiety should be toned down. The meta-
analytic results imply that the valence-based bias in anxiety is a
function of several cognitive processes, including preattentive,
attentional, and postattentive processes.
Because existing models cannot account for the outline of the
findings that emerged from the present meta-analysis, a new the-
oretical framework is in order. We offer a tentative integrative
model (see Figure 1) that incorporates several aspects of previous
models and is consistent with the findings of the present meta-
analysis. Instead of assigning the bias to a malfunction of only one
cognitive process, we propose that anxious individuals may dis-
play abnormal processing patterns at each of four different stages
or in different combinations of these. According to this model, a
preattentive threat evaluation system (PTES) preattentively eval-
uates stimuli in the environment. A stimulus that is tagged with a
high threat value feeds forward into a resource allocation system
(RAS) and triggers a physiological alert state, interruption of
ongoing activity, allocation of processing resources to the stimu-
lus, and a conscious anxious state. These outcomes lead to a set of
strategic processes carried out by a guided threat evaluation system
(GTES). At this stage, assessment of the context of the threat
stimulus, comparison of the present threat with prior learning and
memory, and assessment of the availability of coping resources
take place. If the outcome of this guided threat evaluation results
in a low conscious threat evaluation, a feedback process is trig-
gered that overrides the input emanating from the PTES and
relaxes the alert state imposed by the RAS. If, in contrast, the result
of this guided evaluation corroborates the threat alert invoked by
the PTES, a high state of anxiety is likely to proceed.
Construed in this manner, high-trait anxiety or different anxiety
disorders may stem from (a) a tendency to automatically evaluate
benign or slightly threatening stimuli as high threat; (b) a bias in
the RAS, that is, a tendency to allocate resources even to stimuli
evaluated as only mildly threatening; (c) a tendency to consciously
evaluate alert signals as highly threatening even when context,
prior learning, and available coping resources may indicate the
contrary; or (d) deficiencies in the overriding mechanism, in which
Figure 1. A model of the cognitive mechanisms underlying threat processing.
case even conscious understanding of the irrational aspects of the
threat evaluation cannot terminate the anxious state.
Taking into account the striking finding of a similar bias across
anxious populations, one may speculate that phenotypic differ-
ences between different anxious populations may be determined
by the specific pattern of biases and malfunctions at different
stages of the model. For instance, specific phobia is characterized
by marked and persistent fear that is cued by the presence or
anticipation of a specific object or situation. Exposure to the
phobic stimulus almost invariably provokes an immediate anxiety
response, which in our model may be attributed to a specific bias
in the PTES, followed by an alert response of the RAS. An
additional diagnostic feature of specific phobia is that the person
recognizes that his or her fear is excessive and unreasonable,
which, according to our model, may indicate that the GTES is
functioning properly in specific phobia but that the overriding
mechanism that is expected to relax the alert state imposed by the
RAS may be dysfunctional.
Other anxiety disorders may involve dysfunctions in other
stages of threat processing. For example, Foa and colleagues (Foa,
Feske, Murdock, Kozak, & McCarthy, 1991; Foa, Steketee, &
Rothbaum 1989) have proposed that patients with PTSD selec-
tively attend to trauma-related stimuli because this material is
readily activated in fear templates stored in memory as a conse-
quence of the trauma. They have further proposed that when
trauma-related structures are activated in patients with PTSD,
these might interfere with other cognitive mechanisms that are
required for the integration and assessment of incoming informa-
tion. Thus, in addition to the high threat value being assigned to
trauma-related stimuli in the PTES, and the vigorous alarm reac-
tion enforced by the RAS, the GTES of individuals with PTSD
may fail to integrate contextual, coping resources, and other rele-
vant information, leading to the often reported symptom of disso-
ciated reliving of the trauma.
This tentative multistage model offers an integrative conceptual
framework for thinking about anxiety conditions. Although this
model appears to accommodate many of the central findings in the
field as well as various clinical observations, it awaits validation
by direct experimental testing in future research.
Future Directions
With over 150 studies that have established the existence and
typical magnitude of the threat-related bias in anxious individuals
from different populations and with a variety of experimental
conditions, it appears as if little will be gained from additional
studies of threat-related bias unless these are strongly driven by
theory. What then, should be the future directions for research in
this field? We suggest a number of topics that, in our view,
particularly deserve further investigation.
First, there is a need for more refined investigation of the
different stages of information processing in which anxious and
nonanxious people differ. This calls for new experimental setups
(the emotional spatial cuing task recently introduced to the field is
a good example), for the use of other outcome measures in addition
to manual reaction time and accuracy (e.g., response variability),
and for reliance on technologies that allow one to go beyond
observed behavior in order to index the timing of specific cogni-
tive processes (e.g., eye tracking, event-related potentials [ERPs]).
Second, although the notion that threat-related bias may con-
tribute causally to the development and maintenance of anxiety
largely underlies the impetus for research in the field (e.g., Kaspi,
McNally, & Amir, 1995; MacLeod et al., 1986; Mogg, Bradley, et
al., 1993), empirical support for such a causal link is scarce.
Research efforts toward the establishment of this causal link are of
primary importance. MacLeod, Rutherford, Campbell, Ebsworthy,
and Holker (2002) took a first step in this direction by experimen-
tally inducing differential attentional responses to emotional stim-
uli in nonanxious participants using a modified dot-probe task, and
then examining the impact of the attentional manipulation on
subsequent emotional vulnerability. The results supported the hy-
pothesis that the induction of attentional bias modified emotional
vulnerability, as revealed by participants’ subsequent emotional
reactions in a standardized stress task. However, whether the
processes tapped in this study are similar to those leading to
clinical anxiety, and whether emotional vulnerability may be ad-
equately equated with anxiety, remains open to alternative inter-
A related issue concerns whether threat-related bias in anxious
individuals might be an epiphenomenon of increased anxiety. Only
a few studies have addressed this issue, and they have provided
support for this hypothesis by showing that successful treatment of
anxious patients led to a reduction or even the abolishment of the
threat-related bias initially observed in these patients (e.g., Lundh
& Oest, 2001; Mattia, Heimberg, & Hope, 1993; Mathews, Mogg,
Kentish, & Eysenck, 1995). Clearly, more research is needed on
this critical issue.
Third, the present review reveals a relative paucity in studies of
threat-related bias in anxious and nonanxious children, with some
areas, such as subliminal exposure times, not covered at all.
Particularly missing are longitudinal studies tracing the develop-
ment and the associations over time between threat-related bias
and anxiety. This lack of longitudinal research is surprising, given
that understanding childhood pathways to anxiety can provide a
unique perspective from which to appreciate the initial structure
and function of anxiety-related information processing in anxiety
disorders. In addition, in future cross-sectional studies, it may be
extremely useful to narrow the age range of participants, as chil-
dren’s performance on specific attention tasks varies considerably
with age. For instance, it has been carefully documented that
children 3.5 to 4.5 years of age find the day–night Stroop-like task
extremely difficult (Gerstadt, Hong, & Diamond, 1994). Diamond,
Kirkham, and Amso (2002) further demonstrated that whereas
4-year-olds perform at a chance rate (53% correct) on this task,
4.5-year-olds perform at an almost 80% correct rate, and for 6- to
7-year-olds, the task becomes trivially easy. These findings indi-
cate that around 4 years of age, there exists a sensitive period of
development in attention control functions tapped by this particu-
lar task. Future studies of attentional bias in anxious children
would benefit immensely from well-documented norms on the
relevant cognitive tasks.
Finally, it seems that the time is ripe for research into the neural
substrates of the threat-related bias in anxious individuals. One
example, albeit not using an attention paradigm, is a study by
Straube, Kolassa, Glauer, Mentzel, and Miltner (2004). They used
event-related functional magnetic resonance imaging to assess
brain activation in response to photographs and schematic pictures
depicting angry or neutral facial expressions in participants with
social phobia and in healthy control participants. Straube et al.
found that differences between participants with social phobia and
nonanxious controls in brain responses to socially threatening
faces were most pronounced when facial expression was task
irrelevant, and that the insula played a unique role in the intensive
processing of angry facial expressions by the group with social
phobia. An additional study by Bar-Haim, Lamy, and Glickman
(2005) used event-related brain potentials to study the deployment
of attention to face stimuli with different emotion expressions in
high-anxious and low-anxious participants. The ERP data indi-
cated that threat-related faces elicited faster latencies and greater
amplitudes of early ERP components in high-anxious individuals
than in low-anxious individuals. More studies using different at-
tention paradigms and different measures of brain activity are
needed in order to further elucidate the neural correlates of threat-
related biases in anxiety.
References marked with an asterisk indicate studies included in the
Albano, A. M., Chorpita, B. F., & Barlow, D. H. (2003). Childhood anxiety
disorders. In E. J. Mash & R. A. Barkley (Eds.), Child psychopathology
(pp. 279 –329). New York: Guilford Press.
Algom, D., Chajut, E., & Lev, S. (2004). A rational look at the emotional
Stroop phenomenon: A generic slowdown, not a Stroop effect. Journal
of Experimental Psychology: General, 133, 323–338.
*Amir, N., Elias, J., Klumpp, H., & Przeworski, A. (2003). Attentional bias
to threat in social phobia: Facilitated processing of threat or difficulty
disengaging attention from threat? Behaviour Research and Therapy, 41,
Amir, N., Foa, E. B., & Coles, M. E. (1998). Automatic activation and
strategic avoidance of threat-relevant information in social phobia. Jour-
nal of Abnormal Psychology, 107, 285–290.
*Andrews, T. M., & Anderson, I. M. (1998). Information processing in
anxiety: A pilot study of the effect of manipulating 5-HT function.
Journal of Psychopharmacology, 12, 155–160.
*Asmundson, G. J. G., Sandler, L. S., Wilson, K. G., & Walker, J. R.
(1992). Selective attention toward physical threat in patients with panic
disorder. Journal of Anxiety Disorders, 6, 295–303.
*Asmundson, G. J. G., & Stein, M. B. (1994). Selective processing of
social threat in patients with generalized social phobia: Evaluation using
a dot-probe paradigm. Journal of Anxiety Disorders, 8, 107–117.
Bakermans-Kranenburg, M. J., van IJzendoorn, M. H., & Juffer, F. (2003).
Less is more: Meta-analyses of sensitivity and attachment interventions
in early childhood. Psychological Bulletin, 129, 195–215.
Bar-Haim, Y., Lamy, D., & Glickman, S. (2005). Attentional bias in
anxiety: A behavioral and ERP study. Brain and Cognition, 59, 11–22.
Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New
York: International Universities Press.
Beck, A. T., & Clark, D. A. (1997). An information processing model of
anxiety: Automatic and strategic processes. Behaviour Research and
Therapy, 35, 49 –58.
Beck, A. T., Emery, G., & Greenberg, R. L. (1985). Anxiety disorders and
phobias: A cognitive perspective. New York: Basic Books.
*Beck, J., Freeman, J. B., Shipherd, J. C., Hamblen, J. L., & Lackner, J. M.
(2001). Specificity of Stroop interference in patients with pain and
PTSD. Journal of Abnormal Psychology, 110, 536 –543.
*Becker, E. S., Rinck, M., Margraf, J., & Roth, W. T. (2001). The
emotional Stroop effect in anxiety disorders: General emotionality or
disorder specificity? Journal of Anxiety Disorders, 15, 147–159.
*Beckham, J. C., Lytle, B. L., Vrana, S. R., Hertzberg, M. A., Feldman,
M. E., & Shipley, R. H. (1996). Smoking withdrawal symptoms in
response to a trauma-related stressor among Vietnam combat veterans
with posttraumatic stress disorder. Addictive Behaviors, 21, 93–101.
Bower, G. H. (1981). Mood and memory. American Psychologist, 36,
129 –148.
Bower, G. H. (1987). Commentary on mood and memory. Behaviour
Research and Therapy, 25, 443– 455.
*Bradley, B. P., Mogg, K., Falla, S. J., & Hamilton, L. R. (1998). Atten-
tional bias for threatening facial expressions in anxiety: Manipulation of
stimulus duration. Cognition & Emotion, 12, 737–753.
*Bradley, B. P., Mogg, K., & Millar, N. H. (2000). Covert and overt
orienting of attention to emotional faces in anxiety. Cognition & Emo-
tion, 14, 789 808.
*Bradley, B. P., Mogg, K., Millar, N., Bonham-Carter, C., Fergusson, E.,
Jenkins, J., et al. (1997). Attentional biases for emotional faces. Cogni-
tion & Emotion, 11, 25– 42.
*Bradley, B. P., Mogg, K., Millar, N., & White, J. (1995). Selective
processing of negative information: Effects of clinical anxiety, concur-
rent depression, and awareness. Journal of Abnormal Psychology, 104,
*Bradley, B. P., Mogg, K., White, J., Groom, C., & de Bono, J. (1999).
Attentional bias for emotional faces in generalized anxiety disorder.
British Journal of Clinical Psychology, 38, 267–278.
*Broadbent, D., & Broadbent, M. (1988). Anxiety and attentional bias:
State and trait. Cognition & Emotion, 2, 165–183.
*Broomfield, N. M., & Turpin, G. (2005). Covert and overt attention in
trait anxiety: A cognitive psychophysiological analysis. Biological Psy-
chology, 68, 179 –200.
*Brosschot, J. F., de Ruiter, C., & Kindt, M. (1999). Processing bias in
anxious subjects and repressors, measured by emotional Stroop interfer-
ence and attentional allocation. Personality and Individual Differences,
26, 777–793.
*Bryant, R. A., & Harvey, A. G. (1995). Processing threatening informa-
tion in posttraumatic stress disorder. Journal of Abnormal Psychology,
104, 537–541.
*Buckley, T. C., Blanchard, E. B., & Hickling, E. J. (2002). Automatic and
strategic processing of threat stimuli: A comparison between PTSD,
panic disorder, and nonanxiety controls. Cognitive Therapy and Re-
search, 26, 97–115.
Buckley, T. C., Blanchard, E. B., & Neill, W. T. (2000). Information
processing and PTSD: A review of the empirical literature. Clinical
Psychology Review, 20, 1041–1065.
*Carrigan, M. H., Drobes, D. J., & Randall, C. L. (2004). Attentional bias
and drinking to cope with social anxiety. Psychology of Addictive
Behaviors, 18, 374 –380.
*Carter, C. S., Maddock, R. J., & Magliozzi, J. (1992). Patterns of abnor-
mal processing of emotional information in panic disorder and major
depression. Psychopathology, 25, 65–70.
*Cassiday, K. L., McNally, R. J., & Zeitlin, S. B. (1992). Cognitive
processing of trauma cues in rape victims with post-traumatic stress
disorder. Cognitive Therapy and Research, 16, 283–295.
*Chen, E., Lewin, M. R., & Craske, M. G. (1996). Effects of state anxiety
on selective processing of threatening information. Cognition & Emo-
tion, 10, 225–240.
Clark, D. M., & McManus, F. (2002). Information processing in social
phobia. Biological Psychiatry, 51, 92–100.
Clark, D. M., & Wells, A. (1995). A cognitive model of social phobia. In
M. Liebowitz & R. G. Heimberg (Eds.), Social phobia: Diagnosis,
assessment, and treatment (pp. 69 –93). New York: Guilford Press.
*Constans, J. I., McCloskey, M. S., Vasterling, J. J., Brailey, K., &
Mathews, A. (2004). Suppression of attentional bias in PTSD. Journal of
Abnormal Psychology, 113, 315–323.
Cooper, H., & Hedges, L. V. (1994). The handbook of research synthesis.
New York: Russell Sage Foundation.
Cooper, H. M., & Lindsay, J. J. (1998). Research synthesis and meta-
analysis. In L. Bickman & D. J. Rog (Eds.), Handbook of applied social
research methods (pp. 315–337). Thousand Oaks, CA: Sage.
*Dalgleish, T. (1995). Performance on the emotional Stroop task in groups
of anxious, expert, and control subjects: A comparison of computer and
card presentation formats. Cognition & Emotion, 9, 341–362.
*Dalgleish, T., Moradi, A. R., Taghavi, M. R., Neshat Doost, H. T., &
Yule, W. (2001). An experimental investigation of hypervigilance for
threat in children and adolescents with post-traumatic stress disorder.
Psychological Medicine, 31, 541–547.
*Dalgleish, T., Taghavi, R., Neshat Doost, H., Moradi, A., Canterbury, R.,
& Yule, W. (2003). Patterns of processing bias for emotional informa-
tion across clinical disorders: A comparison of attention, memory, and
prospective cognition in children and adolescents with depression, gen-
eralized anxiety, and posttraumatic stress disorder. Journal of Clinical
Child and Adolescent Psychology, 32, 10 –21.
Dalgleish, T., & Watts, F. N. (1990). Biases of attention and memory in
disorders of anxiety and depression. Clinical Psychology Review, 10,
589 604.
*Dawkins, K., & Furnham, A. (1989). The colour naming of emotional
words. British Journal of Psychology, 80, 383–389.
De Ruiter, C., & Brosschot, J. F. (1994). The emotional Stroop interference
effect in anxiety: Attentional bias or cognitive avoidance? Behaviour
Research and Therapy, 32, 315–319.
*Devineni, T., Blanchard, E. B., Hickling, E. J., & Buckley, T. C. (2004).
Effect of psychological treatment on cognitive bias in motor vehicle
accident-related posttraumatic stress disorder. Journal of Anxiety Disor-
ders, 18, 211–231.
Diamond, A., Kirkham, N., & Amso, D. (2002). Conditions under which
young children can hold two rules in mind and inhibit a prepotent
response. Developmental Psychology, 38, 352–362.
Di Lollo, V., Enns, J. T., & Rensink, R. A. (2000). Competition for
consciousness among visual events: The psychophysics of reentrant
visual processes. Journal of Experimental Psychology: General, 129,
Driver, J. (2001). A selective review of selective attention research from
the past century. British Journal of Psychology, 92, 53–78.
*Dubner, A. E., & Motta, R. W. (1999). Sexually and physically abused
foster care children and posttraumatic stress disorder. Journal of Con-
sulting and Clinical Psychology, 67, 367–373.
Eastwood, J. D., Smilek, D., & Merikle, P. M. (2001). Differential atten-
tional guidance by unattended faces expressing positive and negative
emotion. Perception and Psychophysics, 63, 1004 –1013.
*Egloff, B., & Hock, M. (2000). Interactive effects of state anxiety and trait
anxiety on emotional Stroop interference. Personality and Individual
Differences, 31, 875– 882.
Egloff, B., & Hock, M. (2001). Interactive effects of state anxiety and trait
anxiety on emotional Stroop interference. Personality and Individual
Differences, 31, 875– 882.
*Egloff, B., & Hock, M. (2003). Assessing attention allocation toward
threat-related stimuli: A comparison of the emotional Stroop task and the
attentional probe task. Personality and Individual Differences, 35, 475–
*Ehlers, A., Margraf, J., Davies, S., & Roth, W. T. (1988). Selective
processing of threat cues in subjects with panic attacks. Cognition &
Emotion, 2, 201–219.
Ehrenreich, J. T., & Gross, A. M. (2002). Biased attentional behavior in
childhood anxiety: A review of theory and current empirical investiga-
tion. Clinical Psychology Review, 22, 991–1008.
Eizenman, M., Yu, L. H., Grupp, L., Eizenman, E., Ellenbogen, M.,
Gemar, M., et al. (2003). A naturalistic visual scanning approach to
assess selective attention in major depressive disorder. Psychiatry Re-
search, 118, 117–128.
*Elsesser, K., Sartory, G., & Tackenberg, A. (2004). Attention, heart rate,
and startle response during exposure to trauma-relevant pictures: A
comparison of recent trauma victims and patients with posttraumatic
stress disorder. Journal of Abnormal Psychology, 113, 289 –301.
Eysenck, M. W. (1992). Anxiety: The cognitive perspective. Hove, En-
gland: Psychology Press.
Feehan, M., McGee, R., & Williams, S. M. (1993). Mental health disorders
from age 15 to age 18 years. Journal of the American Academy of Child
and Adolescent Psychiatry, 32, 1118 –1126.
Ferdinand, R. F., & Verhulst, F. C. (1995). Psychopathology from adoles-
cence into young adulthood: An 8-year follow-up study. American
Journal of Psychiatry, 152, 1586 –1594.
*Field, N. P., Classen, C., Butler, L. D., Koopman, C., Zarcone, J., &
Spiegel, D. (2001). Revictimization and information processing in
women survivors of childhood sexual abuse. Journal of Anxiety Disor-
ders, 15, 459 469.
*Foa, E. B., Feske, U., Murdock, T. B., Kozak, M. J., & McCarthy, P. R.
(1991). Processing of threat-related information in rape victims. Journal
of Abnormal Psychology, 100, 156 –162.
*Foa, E. B., Ilai, D., McCarthy, P. R., Shoyer, B., & Murdock, T. (1993).
Information processing in obsessive– compulsive disorder. Cognitive
Therapy and Research, 17, 173–189.
Foa, E. B., & Kozak, M. J. (1986). Emotional processing of fear: Exposure
to corrective information. Psychological Bulletin, 99, 20 –35.
Foa, E. B., Steketee, G., & Rothbaum, B. (1989). Behavioral/cognitive
conceptualizations of posttraumatic stress disorder. Behavior Therapy,
20, 155–176.
*Fox, E. (1993a). Allocation of visual attention and anxiety. Cognition &
Emotion, 7, 207–215.
*Fox, E. (1993b). Attentional bias in anxiety: Selective or not? Behaviour
Research and Therapy, 31, 487– 493.
*Fox, E. (1994). Attentional bias in anxiety: A defective inhibition hy-
pothesis. Cognition & Emotion, 8, 165–195.
*Fox, E. (2002). Processing emotional facial expressions: The role of
anxiety and awareness. Cognitive, Affective and Behavioral Neuro-
science, 2, 52– 63.
Fox, E. (2004). Maintenance or capture of attention in anxiety-related
biases? In J. Yiend & A. M. Mathews (Eds.), Cognition, emotion and
psychopathology: Theoretical, empirical and clinical directions (pp.
86 –105). New York: Cambridge University Press.
Fox, E., Lester, V., Russo, R., Bowles, R. J., & Dutton, K. (2000). Facial
expressions of emotion: Are angry faces detected more efficiently?
Cognition & Emotion, 14, 61–92.
*Fox, E., Russo, R., Bowles, R., & Dutton, K. (2001). Do threatening
stimuli draw or hold visual attention in subclinical anxiety? Journal of
Experimental Psychology: General, 130, 681–700.
*Fox, E., Russo, R., & Dutton, K. (2002). Attentional bias for threat:
Evidence for delayed disengagement from emotional faces. Cognition &
Emotion, 16, 355–379.
Gerstadt, C., Hong, Y., & Diamond, A. (1994). The relationship between
cognition and action: Performance of 3
–7 year old children on a
Stroop-like day–night test. Cognition, 53, 129 –153.
Gilboa Schechtman, E., Foa, E. B., & Amir, N. (1999). Attentional biases
for facial expressions in social phobia: The face-in-the-crowd paradigm.
Cognition & Emotion, 13, 305–318.
*Golombok, S., Mathews, A., MacLeod, C., & Lader, M. (1990). The
effects of diazepam on cognitive processing. Human Psychopharmacol-
ogy: Clinical and Experimental, 5, 143–147.
*Golombok, S., Stavrou, A., Bonn, J., Mogg, K., Critchlow, S., & Rust, J.
(1991). The effects of diazepam on anxiety-related cognition. Cognitive
Therapy and Research, 15, 459 467.
*Gotlib, I. H., Krasnoperova, E., Yue, D. N., & Joormann, J. (2004).
Attentional biases for negative interpersonal stimuli in clinical depres-
sion. Journal of Abnormal Psychology, 113, 121–135.
Green, M., Rogers, P. J., & Elliman, N. A. (1995). Change in affective state
assessed by impaired color-naming of threat-related words. Current
Psychology: Developmental, Learning, Personality, Social, 14, 222–
Hadwin, J. A., Donnelly, N., French, C. C., Richards, A., Watts, A., &
Daley, D. (2003). The influence of children’s self-report trait anxiety and
depression on visual search for emotional faces. Journal of Child Psy-
chology and Psychiatry and Allied Disciplines, 44, 432– 444.
Hansen, C. H., & Hansen, R. D. (1988). Finding the face in the crowd: An
anger superiority effect. Journal of Personality and Social Psychology,
54, 917–924.
*Harvey, A. G., Bryant, R. A., & Rapee, R. M. (1996). Preconscious
processing of threat in posttraumatic stress disorder. Cognitive Therapy
and Research, 20, 613– 623.
Heinrichs, N., & Hofman, S. G. (2001). Information processing in social
phobia: A critical review. Clinical Psychology Review, 21, 751–770.
*Holle, C., Neely, J. H., & Heimberg, R. G. (1997). The effects of blocked
versus random presentation and semantic relatedness of stimulus words
on response to a modified Stroop task among social phobics. Cognitive
Therapy and Research, 21, 681– 697.
*Hope, D. A., Rapee, R. M., Heimberg, R. G., & Dombeck, M. J. (1990).
Representations of the self in social phobia: Vulnerability to social
threat. Cognitive Therapy and Research, 14, 177–189.
*Hopko, D. R., McNeil, D. W., Gleason, P. J., & Rabalais, A. E. (2002).
The emotional Stroop paradigm: Performance as a function of stimulus
properties and self-reported mathematics anxiety. Cognitive Therapy
and Research, 26, 157–166.
*Johnsen, B. H., Thayer, J. F., Laberg, J. C., Wormnes, B., Raadal, M.,
Skaret, E., et al. (2003). Attentional and physiological characteristics of
patients with dental anxiety. Journal of Anxiety Disorders, 17, 75– 87.
*Jones, G. V., Stacey, H., & Martin, M. (2002). Exploring the intensity
paradox in emotional Stroop interference. Cognitive Therapy and Re-
search, 26, 831– 839.
*Kampman, M., Keijsers, G. P. J., Verbraak, M. J. P. M., Naring, G., &
Hoogduin, C. A. L. (2002). The emotional Stroop: A comparison of
panic disorder patients, obsessive–compulsive patients, and normal con-
trols, in two experiments. Journal of Anxiety Disorders, 16, 425– 441.
*Kaspi, S. P., McNally, R. J., & Amir, N. (1995). Cognitive processing of
emotional information in posttraumatic stress disorder. Cognitive Ther-
apy and Research, 19, 433– 444.
*Keogh, E., Dillon, C., Georgiou, G., & Hunt, C. (2001). Selective atten-
tional biases for physical threat in physical anxiety sensitivity. Journal
of Anxiety Disorders, 15, 299 –315.
*Keogh, E., Ellery, D., Hunt, C., & Hannent, I. (2001). Selective atten-
tional bias for pain-related stimuli amongst pain fearful individuals.
Pain, 91, 91–100.
*Kim, H. Y., Lundh, L. G., & Harvey, A. (2002). The enhancement of
video feedback by cognitive preparation in the treatment of social
anxiety: A single-session experiment. Journal of Behavior Therapy and
Experimental Psychiatry, 33, 19 –37.
*Kindt, M., Bierman, D., & Brosschot, J. F. (1997). Cognitive bias in
spider fear and control children: Assessment of emotional interference
by a card format and a single-trial format of the Stroop task. Journal of
Experimental Child Psychology, 66, 163–179.
*Kindt, M., Bogels, S., & Morren, M. (2003). Processing bias in children
with separation anxiety disorder, social phobia and generalized anxiety
disorder. Behaviour Change, 20, 143–150.
*Kindt, M., & Brosschot, J. F. (1997). Phobia-related cognitive bias for
pictorial and linguistic stimuli. Journal of Abnormal Psychology, 106,
644 648.
*Kindt, M., & Brosschot, J. F. (1998). Stability of cognitive bias for threat
cues in phobia. Journal of Psychopathology and Behavioral Assessment,
20, 351–367.
*Kindt, M., & Brosschot, J. F. (1999). Cognitive bias in spider-phobic
children: Comparison of a pictorial and a linguistic spider Stroop.
Journal of Psychopathology and Behavioral Assessment, 21, 207–220.
*Kindt, M., Brosschot, J. F., & Everaerd, W. (1997). Cognitive processing
bias of children in a real life stress situation and a neutral situation.
Journal of Experimental Child Psychology, 64,