Attentional bias
Recurring concerns steer what we notice.
What it means
Attentional bias is the tendency for perception and attention to be drawn preferentially toward certain categories of information, typically those that are emotionally or motivationally significant to the individual, at the expense of equally available alternatives. The proposed mechanism combines rapid capture with difficulty disengaging: salient or self-relevant stimuli pull attention and then hold it, so a person's preoccupations filter what they perceive, which may feed a self-reinforcing loop in anxiety, depression, addiction and chronic pain. The evidence is stronger for the idea than for its measurement. An early meta-analysis put the threat bias in anxious groups at a moderate d = 0.45, but the reaction-time difference scores it rests on have close to zero reliability, and a later meta-analysis of trial baselines found no threat bias in clinically anxious patients. Attention-bias-modification training, premised on the bias causing symptoms, yields small and inconsistent benefits. Read attentional bias as a description of how attention behaves in the moment, not as a stable quantity you can score in a person.
How it works
Attentional bias is usually decomposed into three separable components. Facilitated engagement means significant stimuli win the race early: the threat is found faster than a neutral item in the same display. Difficulty disengaging means that once attention lands it is slow to leave, and much of the measured effect comes from this rather than from faster detection. Attentional avoidance is the late move away: after an initial orienting, people often look pointedly elsewhere, which is why the direction of the bias flips with how long a stimulus is shown. The components run on different clocks, so a 100 ms presentation and a 1,000 ms one can produce opposite group-level results. The structure is real — it shows up most clearly where attention is tracked directly rather than inferred from a key press — but as the next section explains, it is not something a reaction-time score can resolve in an individual.
What the evidence shows
The foundational demonstration is MacLeod, Mathews and Tata (1986): anxious patients detected a probe faster when it replaced a threat word than a neutral one, while controls showed the opposite. Two decades of work followed, and a 2007 meta-analysis of 172 studies put the threat bias in anxious groups at d = 0.45, present across paradigms and even for stimuli presented outside awareness. Then the picture darkened. A 2019 meta-analysis pooled baseline dot-probe scores from 1,005 patients with clinical anxiety or PTSD across 13 treatment trials and found a mean bias of 1.8 ms, indistinguishable from zero, with Bayesian analysis favouring the null. The two are hard to reconcile. Bar-Haim pooled several paradigms; Kruijt is dot-probe only, in patients who had enrolled in treatment trials. Publication bias in the earlier literature is the most-cited explanation, and probably part of it, but the paradigm and sample differences do real work too. The honest reading is that the group-level effect is far less certain than the textbook version suggests.
Why the measure is the weak link
A dot-probe score is the difference between two reaction times, each already a noisy average. Subtracting one from the other cancels the reliable part, a person's general speed, and keeps the error. The result is the reliability paradox: a task that yields a robust average effect and near-useless individual scores. A 2025 study ran 36 variations of the emotional dot-probe on 9,600 people and found no version with adequate split-half reliability, while median reaction times in the same data were reliable at .98. The problem is the difference score, not the measurement. The consequence is blunt: an unreliable measure caps the correlation it can produce, so correlating individual bias scores with anxiety was never going to work. Much of the literature built on those scores was mostly measuring noise.
Using it in practice
The surviving practical claim is narrow. Attention-bias-modification training, which repeatedly steers probes away from threat, was expected to cut symptoms if the bias caused them. A 2020 network meta-analysis of 65 anxiety trials found attention training helped only in post-hoc analyses excluding PTSD studies, and across the whole 85-trial review only four trials were at low risk of bias on all six domains assessed; interpretation-bias training fared better. In addiction, bias and self-reported craving correlate at about r = .19, under 4 percent shared variance, and shared variance rises to about 13 percent when attention is measured directly, by eye movements or event-related potentials, rather than inferred from reaction times. So use attentional bias to explain why someone cannot stop scanning the cigarette display, not to screen or score them. If you need a number, watch where the eyes go rather than how fast a key gets pressed.
Examples
An anxious person's eyes snap to angry faces in a crowd faster than to friendly ones.
Someone three days into quitting smoking clocks the newsagent's cigarette display from across the road and keeps glancing back, while the friend beside them never registers it at all.
After a round of layoffs, an employee reads threat into every neutral message from HR and spends the afternoon watching their manager's door instead of working.
A patient with chronic back pain reads a hospital leaflet and snags on every mention of damage and surgery, and has re-read the paragraph on surgical risk twice before noticing the recovery figures in the next column at all.
A new parent surfaces from deep sleep at one thin cry from the next room but sleeps through a passing siren that would have woken them a month earlier.
First described in Studied via the emotional Stroop and dot-probe tasks.
Key references
- Xu, I., Passell, E., Strong, R. W., Grinspoon, E., Jung, L., Wilmer, J. B., & Germine, L. T. (2025). No evidence of reliability across 36 variations of the emotional dot-probe task in 9,600 participants. Clinical Psychological Science, 13(2), 261-277. doi.org/10.1177/21677026241253826
- Fodor, L. A., Georgescu, R., Cuijpers, P., Szamoskozi, S., David, D., Furukawa, T. A., & Cristea, I. A. (2020). Efficacy of cognitive bias modification interventions in anxiety and depressive disorders: A systematic review and network meta-analysis. The Lancet Psychiatry, 7(6), 506-514. doi.org/10.1016/S2215-0366(20)30130-9
- Kruijt, A.-W., Parsons, S., & Fox, E. (2019). A meta-analysis of bias at baseline in RCTs of attention bias modification: No evidence for dot-probe bias towards threat in clinical anxiety and PTSD. Journal of Abnormal Psychology, 128(6), 563-573. doi.org/10.1037/abn0000406
- Field, M., Munafo, M. R., & Franken, I. H. A. (2009). A meta-analytic investigation of the relationship between attentional bias and subjective craving in substance abuse. Psychological Bulletin, 135(4), 589-607. doi.org/10.1037/a0015843
- Bar-Haim, Y., Lamy, D., Pergamin, L., Bakermans-Kranenburg, M. J., & van IJzendoorn, M. H. (2007). Threat-related attentional bias in anxious and nonanxious individuals: A meta-analytic study. Psychological Bulletin, 133(1), 1-24. doi.org/10.1037/0033-2909.133.1.1
- MacLeod, C., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorders. Journal of Abnormal Psychology, 95(1), 15-20. doi.org/10.1037/0021-843X.95.1.15