Vigilance decline stems from bias and lapses, not sensitivity loss

Vigilance has long been one of psychology’s most stubborn puzzles. Nearly eight decades after Norman Mackworth first documented that human observers watching a radar-like display steadily lose their ability to detect faint signals, scientists have continued to argue about what actually goes wrong inside the mind as minutes tick by. Does the observer’s perceptual system genuinely become worse at telling signal from noise, or does something more subtle—criterion drift, wandering thoughts, disengagement—masquerade as sensory failure? A new study published in Attention, Perception, & Psychophysics delivers what its authors describe as decisive evidence on this question, and the answer points away from the eye and toward the decision.

The research, led by Jason S. McCarley of Oregon State University together with Shannon P. Gyles, Kayla R. Hankey of Old Dominion University, Fernando Muñoz Gomez Andrade, and Yusuke Yamani, deconstructed the vigilance decrement—the gradual decline in detection performance over the course of a sustained attention task—into its candidate mechanistic components. Rather than measuring a single hit rate and inferring what changed, the team used psychometric curve analysis, a technique borrowed from classical psychophysics that models how detection probability varies across a continuum of stimulus strengths. By fitting these curves over time, the researchers could mathematically separate four distinct contributors to performance: the observer’s response bias, their underlying perceptual sensitivity, their rate of mental lapses, and their tendency to guess positively.

The logic of the decomposition is grounded in signal detection theory, the framework introduced by David Green and John Swets in 1966 that treats every detection decision as a judgment about whether noisy internal evidence crossed a criterion. In this framework, sensitivity reflects how separable the distributions of evidence are for signal and noise trials; bias reflects where the observer places their decision threshold; and additional parameters can capture trials on which attention momentarily fails entirely, producing lapses, or on which the observer responds affirmatively without processing the stimulus at all, producing guesses. Each of these components leaves a characteristic fingerprint on the shape and placement of the psychometric function, which means that with enough data, competing theories of the vigilance decrement can be tested against one another rather than merely asserted.

To generate those data, the team recruited 366 participants for a demanding vigilant monitoring task. On each trial, observers viewed a pair of small visual probes whose spatial separation varied from trial to trial. The observer’s job was to report the occasional trials on which the separation between the two probes exceeded a criterion value—an infrequent target event embedded in a stream of near-threshold discriminations. Crucially, the stimuli were masked by dynamic visual noise, and trials arrived at a punishing forced pace of 40 per minute. These conditions were deliberately engineered to elevate processing demands and to maximize the chance that a genuine sensitivity loss, if one exists, would appear. If vigilance decrements are caused by fatigue-like degradation of the perceptual machinery, this was the environment in which that degradation should be most visible.

The results were unambiguous in one direction and surprising in another. Bayesian model comparisons, which weigh competing parameterizations of the psychometric functions against one another, revealed decisive evidence for three time-dependent changes: observers’ response criteria became progressively more conservative, meaning they became less willing to call a trial a target even when the perceptual evidence was the same; their mental lapse rates increased, meaning they were more likely to fail to process a trial at all; and their positive guess rate declined, meaning they made fewer affirmative responses on trials where they were effectively answering at random. But the model comparisons gave evidence against the one change that many theorists would have predicted: sensitivity—the fundamental ability to discriminate signal from noise—did not deteriorate over time.

This pattern of findings is consequential because the field has been split for decades. One influential tradition, rooted in resource theories of attention, holds that the vigilance decrement reflects an exhaustion of the cognitive resources needed for effortful perceptual processing; on this view, sensitivity should decline with time on task, and some studies using image-degraded displays have seemed to show exactly that. A competing tradition, championed by researchers including David Thomson and colleagues, has argued that the apparent sensitivity losses in modern vigilance tasks may be statistical artifacts, and that shifts in response criterion combined with mind wandering offer a more parsimonious explanation. The new findings land firmly in the second camp, but with greater precision than before, because the psychometric approach does not rely on the coarse hit-and-false-alarm measures that have historically made the debate difficult to settle.

Perhaps the most provocative part of the paper concerns a reanalysis of earlier work. A previous high-profile study of the vigilance decrement had reported evidence consistent with a sensitivity loss, a result that seemed to support the resource-depletion account. When the team reanalyzed those data with correctly specified psychometric models, the apparent sensitivity loss dissolved—it turned out to be a product of model misspecification rather than a real property of the observers. In other words, an earlier generation of conclusions about the waning of perception itself may have been, at least in part, a lesson about the importance of choosing the right statistical model. The finding serves as a cautionary tale for any researcher drawing mechanistic conclusions from aggregated accuracy rates.

The theoretical implications reach into everyday life and high-stakes workplaces alike. Vigilance decrements plague airport security screeners, air traffic controllers, radiologists scanning medical images, and operators monitoring automated systems—any context in which a human must remain alert for rare targets over long stretches. If the decrement reflects a loss of perceptual sensitivity, remedies might emphasize restoring sensory acuity or arousal. But if, as the new results suggest, the decrement instead reflects a drift toward conservative responding and an increasing frequency of mindless, unprocessed trials, then interventions should target motivation, goal maintenance, and engagement. Techniques that periodically reactivate the task goal, or that reset the observer’s criterion, may prove more effective than attempts to “recharge” a depleted perceptual system. The study’s authors note that mind wandering—the spontaneous drift of thought away from the task—has been shown in meta-analytic reviews to increase steadily in frequency over time on task, and rising lapse rates fit naturally with that picture.

The scale and rigor of the study strengthen its case. With 366 participants, a preregistered experimental design, and analytic code and data openly shared on the Open Science Framework, the work meets contemporary standards for reproducibility that the vigilance literature, like much of psychology, has sometimes struggled to satisfy. The Bayesian hierarchical modeling approach allowed the researchers to estimate parameters at both the individual and group levels, quantifying evidence for or against each mechanistic component rather than relying on null hypothesis significance tests that can conflate model failure with parameter stasis. The authors also acknowledge the lineage of their approach: an earlier online vigilance study by members of the same team had already hinted that bias, lapse rate, and guess rate—rather than sensitivity—drive the decrement, and a commentary published in Psychological Science raised generalizability questions that the new, more demanding task was designed to answer.

Skeptics may point out that even a task wrapped in dynamic noise and run at a rate of 40 events per minute is still a laboratory abstraction, and that real-world vigilance settings involve memory loads, stress, and motivation that the present design did not manipulate. The authors themselves situate their results within that broader literature, noting that memory load and event rate have both been shown to modulate sensitivity decrements under some conditions, and that the classic image-degradation findings in Science in the 1980s demonstrated rapid sensitivity losses under specific sensory conditions. The new claim is not that sensitivity can never fall during sustained attention, but that the canonical vigilance decrement observed in standard monitoring tasks does not require, and in this case did not show, any loss of signal-noise discriminability. The phenomenon that has fascinated and frustrated applied psychologists since 1948, on this evidence, is fundamentally a story about decisions and disengagement rather than about perception failing.

For a phenomenon first documented in the era of vacuum-tube radar, the vigilance decrement has proven remarkably durable as a scientific puzzle, and this study adds a decisive datapoint to a debate spanning more than 75 years. What fades over time on a monitoring task, the work suggests, is not the observer’s ability to see the signal but their willingness to commit to seeing it—and their grip on the task itself. As automation continues to push humans into the role of supervisory monitors, understanding that distinction may matter not just for theory but for designing systems that keep the human in the loop genuinely attentive.