The study, led by Sarah Prieto of Brown University’s Alpert Medical School and the Butler Hospital Memory and Aging Program, together with colleagues at Penn State’s Center for Healthy Aging, is part of a growing movement in cognitive aging research that exploits the smartphone as a laboratory in miniature. Participants, aged 60 to 81, were mailed Motorola smartphones preloaded with the Mobile Monitoring of Cognitive Change (M2C2) app. Before each of up to 24 testing sessions, they rated their current stress on a scale from 0 to 100. Then they completed three brief tasks—each roughly a minute long—tapping processing speed, visual working memory, and episodic memory. Adherence was strikingly high: participants completed an average of 95.6 percent of scheduled sessions, a level of compliance rarely achievable in traditional lab-based designs.

The conceptual shift at the heart of the study is the distinction between between-person and within-person effects. Most prior research on stress and cognition has compared groups of people: those who report chronically high stress versus those who report low stress. But such comparisons can obscure something important—how a person’s performance changes when their stress deviates from their own baseline. A person who is generally anxious but feeling unusually calm on a given afternoon may perform very differently from someone who is normally relaxed but suddenly rattled. By computing each participant’s average stress across all sessions and then flagging moments when ratings exceeded that personal mean by half a standard deviation, the team could isolate these momentary deviations and ask whether they predicted moment-to-moment changes in cognitive performance.

The headline result concerned processing speed. On the Symbol Match task—a timed two-alternative forced-choice paradigm in which participants judge whether symbol pairs match—higher-than-usual stress was associated with faster reaction times, with an unstandardized estimate of −3.82 milliseconds per unit of within-person stress (p < .001). In other words, when participants felt more stressed than was typical for them, they responded more quickly. But this effect was not static. A significant interaction between stress and session number (p = .003) revealed that the speed advantage diminished as the study progressed. Johnson–Neyman analyses pinpointed the boundary: the stress-related speedup was statistically significant only during the first 12.6 sessions, after which the association faded to null. By the final sessions, the pattern had reversed direction, with elevated stress trending toward slower responses.

Accuracy told a complementary story. On the same Symbol Match task, within-person stress increases were associated with a small but significant reduction in accuracy (Est. = −0.025, p = .042), and this negative association also weakened across sessions. Taken together, the reaction time and accuracy findings suggest a classic speed-accuracy trade-off: early in the study, stressed moments pushed participants toward faster but slightly less careful responding. The authors interpret this through the lens of the Yerkes-Dodson framework, the century-old principle that performance improves with arousal up to an optimal point and then deteriorates. Mild, everyday stress in a cognitively healthy sample may sit on the beneficial side of that inverted-U curve, mobilizing attentional resources just enough to speed responding without yet degrading it.

Equally notable is what did not change. Working memory, measured with the Color Shapes task—a visual change-detection paradigm requiring participants to detect color changes in shapes across sequential presentations, scored using signal-detection discriminability (d′)—showed no association with within-person stress, between-person stress, or their interaction with session number. Episodic memory, assessed with the Prices task, in which participants incidentally encoded grocery item–price pairs and later completed delayed forced-choice recognition trials, was similarly unaffected. The authors offer plausible neurobiological explanations. Stress-related impairments in working memory are most often documented under acute, high-intensity stress, likely reflecting the prefrontal cortex’s sensitivity to sharp stress responses; the mild, naturalistic fluctuations captured here may simply have been too weak to disrupt prefrontal function. Episodic memory disruption, meanwhile, tends to emerge after chronic or severe stress, when sustained cortisol elevations can damage the metabolically demanding hippocampus through oxidative injury and disrupted neuronal connectivity—conditions that eight days of ordinary daily stress in healthy older adults are unlikely to produce.

The timing of stress relative to memory processing phases may also matter. Stress immediately before or during encoding can enhance or diminish memory formation depending on the emotional salience of the material, whereas stress during consolidation or retrieval is more likely to impair performance. Because the study measured stress only at the moment of testing, it could not disentangle these phase-specific effects—a limitation the authors acknowledge. They also note that their single-item stress rating, while efficient for high-frequency sampling, cannot capture the complexity of stress as a construct, including chronic stress burden, day-to-day fluctuations, or cumulative exposure over months and years.

Methodological rigor was evident throughout. The team used linear mixed-effects models with an autoregressive correlation structure to account for the fact that observations closer in time are more correlated than those further apart. Covariates included session number, time of day, age, education, and sex, with a binary indicator distinguishing the first session from all subsequent ones to control for practice effects—a well-documented phenomenon in repeated cognitive testing. Extreme reaction times exceeding three standard deviations above the sample mean were winsorized rather than excluded, preserving observations while limiting their influence; sensitivity analyses excluding these values entirely yielded the same significant stress-by-session interaction (β = 0.16, p = .015). Alternative thresholds for defining elevated stress—at 1.0 and 1.5 standard deviations above each person’s mean—produced consistent patterns, and variance inflation factors remained well below the conventional threshold of 5, ruling out problematic multicollinearity.

The practical implications extend to how digital cognitive assessments are interpreted in research and clinical settings. As smartphone-based testing becomes a mainstream tool for early detection of cognitive decline and for monitoring in preclinical Alzheimer’s disease trials, unmeasured contextual factors like momentary stress could mask or exaggerate true cognitive deficits. A participant who happens to be stressed during a remote assessment might appear faster or slower than their genuine ability warrants, potentially complicating diagnostic conclusions. The authors argue that incorporating real-time stress measurement—or objective alternatives such as passive sensing and physiological metrics—into digital cognitive platforms could sharpen the sensitivity of early screening and improve the ecological validity of remote monitoring. Prior work has already shown that sleep quality and depressive symptoms, assessed via ecological momentary assessment, meaningfully shape next-day cognition; stress now joins that list of contextual variables worth tracking.

The study also carries a broader message about what variability means. Previous research has shown that people with mild cognitive impairment exhibit greater session-to-session variability in processing speed and visual short-term memory binding than cognitively unimpaired peers, raising the possibility that intraindividual variability itself is an early marker of cognitive vulnerability. But variability has causes, and stress is one of them. If a spike in daily stress produces a temporary shift in processing speed, some of the variability observed in high-frequency digital assessments may reflect transient emotional states rather than underlying neural change. Disentangling these sources will be essential if smartphone-based cognitive monitoring is to fulfill its promise of detecting subtle decline earlier than traditional clinic-based testing. The authors call for future work extending this approach to individuals with mild cognitive impairment, examining lagged effects of stress on later performance, and diversifying samples beyond the predominantly White, highly educated cohort studied here. For now, the findings offer a vivid demonstration that the aging brain’s performance is not a fixed trait but a dynamic process, shaped moment by moment by the ebb and flow of everyday stress.

Subject of Research: Within-person stress fluctuations and cognitive performance in smartphone-based remote cognitive assessment of older adults

Article Title: The impact of stress on cognitive performance in remote smartphone-based assessments: exploring within-person variability

Article References: The impact of stress on cognitive performance in remote smartphone-based assessments: exploring within-person variability. (n.d.). https://doi.org/10.3389/fdgth.2026.1878732

Image Credits: AI Generated

DOI: 10.3389/fdgth.2026.1878732

Keywords: stress, cognitive performance, smartphone assessment, older adults, processing speed, working memory, episodic memory, within-person variability, digital health, M2C2, cognitive aging, ecological momentary assessment