Cited by 1,049Anxiety, Depression, Psychometrics, Treatment, Cognitive ProcessesVisual Attention and Saliency DetectionNeural and Behavioral Psychology Studies

The DOI matches its Crossref record; 144 of the 144 references it cites were found in a registry or an index; none of the 103 sources its claims cite is checked for support yet.

Eye tracking of attention in the affective disorders: A meta-analytic review and synthesis

Open access
Armstrong, Thomas, Olatunji, Bunmi O.·2012·Clinical Psychology Review

Abstract

A large body of research has demonstrated that affective disorders are characterized by attentional biases for emotional stimuli. However, this research relies heavily on manual reaction time (RT) measures that cannot fully delineate the time course and components of attentional bias. Eye tracking technology, which allows relatively direct and continuous measurement of overt visual attention, may provide an important supplement to RT measures. This article reviews eye tracking research on anxiety and depression, evaluating the experimental paradigms and eye movement indicators used to study attentional biases. Also included is a meta-analysis of extant eye tracking research (33 experiments; N=1579) on both anxiety and depression. Relative to controls, anxious individuals showed increased vigilance for threat during free viewing and visual search, and showed difficulty disengaging from threat in visual search tasks, but not during free viewing. In contrast, depressed individuals were not characterized by vigilance for threat during free viewing, but were characterized by reduced orienting to positive stimuli, as well as reduced maintenance of gaze on positive stimuli and increased maintenance of gaze on dysphoric stimuli. Implications of these findings for theoretical accounts of attentional bias in anxiety and depression are discussed, and avenues for future research using eye-tracking technology are outlined.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This meta-analytic review synthesizes 33 eye-tracking experiments (N = 1579) on attentional biases in anxiety and depression. Anxiety was linked to increased vigilance for threat, while depression was linked to reduced orienting to and maintenance of positive stimuli and increased maintenance of dysphoric stimuli.

Key findings

  • Anxious individuals showed an orienting bias toward threat in free viewing compared with non-anxious individuals (k = 20; g = .47, p < .001, CI = .25, .69; FSN = 203), but no significant orienting bias toward pleasant stimuli (k = 11; g = .11, p = .38).
  • Depressed individuals did not show an orienting bias toward threat (k = 4; g = −.01, p = .95), but showed reduced orienting toward pleasant stimuli (k = 6; g = −.24, p < .05, CI = −.47, −.004; FSN = 6).
  • In free viewing, anxious individuals did not differ overall in initial maintenance of gaze on threat (k = 16; g = −.17, p = .26), though spider phobia showed decreased maintenance (k = 3; g = −1.02, p < .001) and PTSD showed marginally increased maintenance (k = 2; g = .60, p = .05).
  • In visual search, anxious individuals fixated threatening targets faster (k = 4; g = −.59, p < .04) and maintained attention longer on threatening distractors (k = 5; g = .54, p < .01); depressed individuals showed longer extended maintenance of gaze on dysphoric stimuli (k = 6; g = .46, p < .01) and less on positive stimuli (k = 6; g = −.80, p < .001).

Methods

  • PsycINFO and PubMed were searched for eye-tracking studies of anxiety or depression in adults; inclusion required eye tracking (video-based infrared or electro-oculography), adult anxiety/depression, and simultaneous emotional and neutral stimuli in regular arrays competing for attention.
  • The meta-analysis included 33 experiments with 1579 participants (anxious N = 563; non-anxious N = 532; depressed N = 162; non-depressed N = 257; unselected N = 65); 28 experiments used free viewing (N = 1337) and 5 used visual search in anxiety (N = 242).
  • Effect sizes used Hedges’s g with random-effects models; heterogeneity was tested with Q and I², moderators with categorical and meta-regression analyses, and publication bias with fail-safe N.

Limitations

  • The paper states that eye tracking measures overt attention only and may miss covert attentional allocation, especially for the maintenance hypothesis; free viewing studies may fail to observe maintenance biases if they occur covertly before initial fixation.
  • The paper states that several analyses relied on small numbers of studies, including depression orienting (k = 4), depression positive orienting (k = 6), depression extended maintenance (k = 6), and visual search (k = 5, mostly spider phobia k = 4), limiting generalizability.
  • The paper states that few studies examined threat beyond 3 seconds, so the full time course of attentional bias could not be realized; evident from the text, anxiety analyses showed significant heterogeneity and moderator analyses provided limited explanation.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.cpr.2012.09.004

Cites

144 of 144 references verified0 citation errors · 0 not checkable

Statistics

19 tests recomputed · 1 with the reported and recomputed p on opposite sides of .05

Verification

Claims the paper makes

0 of 103 cited sources checked · 0 supported · 0 flagged

Show all 132 claims

  • It may state two separate claims as one. 56
  • Probably not a substantial claim. 33
81 of 81
  • The paper relies on this without support.

    Eye tracking has the potential to address enduring questions in research on attentional biases.

    “Given the potential of eye tracking to address enduring questions in research on attentional biases, a meta-analysis was conducted on the extant literature.”
  • The paper relies on this without support.

    Threatening stimuli, as danger cues, likely occur on a continuum of urgency.

    “Threatening stimuli, as danger cues, likely occur on a continuum of urgency.”
  • The paper relies on this without support.

    In free viewing tasks, participants are not required to resist orienting to threat.

    “In free viewing tasks, participants are not required to resist orienting to threat”
3 of 3

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Statistics that don’t add up

Each reported p-value recomputed from its test statistic and degrees of freedom.

  • Inconsistent p-value across alpha = .05: Q (4) = 4.54, p < .05

    Major

    The paper reports p < .05 here; recomputing from the test statistic and its degrees of freedom gives p = .338. The two fall on opposite sides of alpha = .05. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p < .05 · recomputed p = .338

  • Inconsistent p-value: Q (5) = .66, p = .97

    Minor

    The paper reports p = .97 here; recomputing from the test statistic and its degrees of freedom gives p = .985. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .97 · recomputed p = .985

  • Inconsistent p-value: Q (3) = .55, p = .95

    Minor

    The paper reports p = .95 here; recomputing from the test statistic and its degrees of freedom gives p = .908. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .95 · recomputed p = .908

  • Inconsistent p-value: Q (5) = 4.75, p = .55

    Minor

    The paper reports p = .55 here; recomputing from the test statistic and its degrees of freedom gives p = .447. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .55 · recomputed p = .447

  • Inconsistent p-value: Q (1) = 3.54, p = .05

    Minor

    The paper reports p = .05 here; recomputing from the test statistic and its degrees of freedom gives p = .060. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .05 · recomputed p = .060

  • Inconsistent p-value: Q (2) = .05, p = .79

    Minor

    The paper reports p = .79 here; recomputing from the test statistic and its degrees of freedom gives p = .975. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .79 · recomputed p = .975

6 of 6

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Doesn’t add upQ(4) = 4.54reported p < .05 · recomputed p = .338The two fall on opposite sides of .05

    “Q (4) = 4.54, p < .05”

  • Doesn’t add upQ(3) = 0.55reported p = .95 · recomputed p = .908

    “Q (3) = .55, p = .95”

  • Doesn’t add upQ(5) = 4.75reported p = .55 · recomputed p = .447

    “Q (5) = 4.75, p = .55”

  • Doesn’t add upQ(1) = 3.54reported p = .05 · recomputed p = .060

    “Q (1) = 3.54, p = .05”

  • Doesn’t add upQ(2) = 0.05reported p = .79 · recomputed p = .975

    “Q (2) = .05, p = .79”

  • Doesn’t add upQ(5) = 0.66reported p = .97 · recomputed p = .985

    “Q (5) = .66, p = .97”

  • ConsistentQ(19) = 47.09reported p < .001 · recomputed p < .001

    “Q (19) = 47.09, p < .001”

  • ConsistentQ(10) = 17.66reported p = .06 · recomputed p = .061

    “Q (10) = 17.66, p = .06”

  • ConsistentQ(1) = 6.72reported p = .01 · recomputed p = .010

    “Q (1) = 6.72, p = .01”

  • ConsistentQ(5) = 8.78reported p = .12 · recomputed p = .118

    “Q (5) = 8.78, p = .12”

  • ConsistentQ(15) = 58.08reported p < .001 · recomputed p < .001

    “Q (15) = 58.08, p < .001”

  • ConsistentQ(4) = 24.13reported p < .001 · recomputed p < .001

    “Q (4) = 24.13, p < .001”

  • ConsistentQ(4) = 23.45reported p < .001 · recomputed p < .001

    “Q (4) = 23.45, p < .001”

  • ConsistentQ(1) = 0.1reported p = .76 · recomputed p = .752

    “Q (1) = .10, p = .76”

  • ConsistentQ(5) = 1.9reported p = .86 · recomputed p = .863

    “Q (5) = 1.9, p = .86”

  • ConsistentQ(5) = 9.77reported p = .08 · recomputed p = .082

    “Q (5) = 9.77, p = .08”

  • ConsistentQ(1) = 4.4reported p < .05 · recomputed p = .036

    “Q (1) = 4.40, p < .05”

  • ConsistentQ(3) = 10.89reported p < .02 · recomputed p = .012

    “Q (3) = 10.89, p < .02”

  • ConsistentQ(4) = 8.28reported p = .08 · recomputed p = .082

    “Q (4) = 8.28, p = .08”

19 of 19

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Incorrect heterogeneity degrees of freedom for social anxiety studies

    Minor

    A heterogeneity Q test across seven studies has k − 1 = 6 degrees of freedom, not the reported 4.

    Reported 4 · recomputed 6

  • Discussion calls a nonsignificant overall result marginal

    Minor

    The discussion describes an overall marginal tendency to avoid maintaining gaze on threat, but the results state that the overall group difference was not significant (p = .26).

  • PTSD confidence interval is inconsistent with its effect size

    Minor

    The two reported confidence limits have a midpoint of −.60 and do not contain the reported effect size of +.60.

    Reported 0.6 · recomputed -0.6

  • Incorrect heterogeneity degrees of freedom for OCD studies

    Minor

    A heterogeneity Q test across two studies has k − 1 = 1 degree of freedom, not the reported 4.

    Reported 4 · recomputed 1

  • Dysphoric-orienting confidence interval is not centered on its effect size

    Minor

    The reported confidence limits have a midpoint of .045, not the reported Hedges's g of .18. This difference exceeds rounding to the printed precision.

    Reported 0.18 · recomputed 0.045

  • Conflicting counts of orienting-variable types

    Minor

    The methods count 17 event-related and 3 epoch-related vigilance analyses, while the results count 16 and 4, respectively, for the 20-study anxiety orienting analysis.

6 of 6

Citations

Cites

Flagged first, then by citations

Show all 140 references

Cited by

Sorted by citations

Items 601–700 of 1,000

Metadata

Authors
Armstrong, T., & Olatunji, B. O.
Affiliations
Vanderbilt University (US)
Corresponding author
Armstrong, Thomas
Venue
Clinical Psychology Review
Publisher
Elsevier BV
Published
2012-12
Volume
32
Issue
8
Pages
704-723
ISSN
0272-7358 (ISSN-L)
OpenAlex
W2104609554
Cited by
1,049
References
140
Citations by year
2026: 71 · 2025: 80 · 2024: 69 · 2023: 87 · 2022: 85 · 2021: 105 · 2020: 148 · 2019: 92 · 2018: 84 · 2017: 54 · 2016: 58 · 2015: 56 · 2014: 39 · 2013: 16 · 2012: 1
FWCI
5.24
Type
journal article
Language
en
Open access
Locations
  • Clinical Psychology Review (publishedVersion)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (acceptedVersion)
  • PubMed Central (acceptedVersion)
Funders
National Institute of Mental Health (F31MH087018)
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Anxiety, Depression, Psychometrics, Treatment, Cognitive Processes
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
Keywords
eye tracking, attentional bias, mood disorders, anxiety, depression, threat vigilance, visual search, free viewing, emotional stimuli
MeSH terms
Anxiety Disorders, Arousal, Attention, Depressive Disorder, Emotions, Eye Movements, Facial Expression, Fixation, Ocular, Humans, Pattern Recognition, Visual, Reaction Time

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review