Cognitive Fatigue Destabilizes Economic Decision Making Preferences and Strategies

The study investigates how cognitive fatigue—arising from sustained cognitively demanding activity—affects economic decision making. Prior work shows that fatigue can diminish motivation, increase distractibility, alter information processing, and worsen mood, potentially degrading performance and health-related decisions. Economic uncertainty comprises risk (known probabilities) and ambiguity (unknown probabilities). Beyond preferences for uncertainty, decision makers may use different informational strategies: maximizing (computing relative expected value, rEV) versus satisficing (using visually available probability of winning, pWIN). The hypothesis was that cognitive fatigue would increase reliance on less effortful satisficing strategies and reduce maximizing, due to compromised top-down control and aversion to further cognitive effort. The study examines whether cognitive fatigue alters uncertainty preferences and/or choice strategies across gains and losses using incentive-compatible tasks and a between-subjects fatigue manipulation.

Background literature indicates that cognitive fatigue is common and associated with diminished motivation, increased distractibility, altered information processing, and mood changes, which can impair performance (e.g., error detection failures and willingness to take chances). Economic theory distinguishes risk (known probabilities) from ambiguity (unknown probabilities). Cognitive fatigue has been linked to compromised top-down control with relative sparing of automatic processes, and to aversion to additional effort, suggesting shifts toward less effortful strategies. Prior work also shows that satisficing strategies are less cognitively costly than maximizing strategies, though maximizing yields higher expected outcomes. Neurobiological evidence suggests partial dissociation between risk and ambiguity processing. Studies of time-on-task and fatigue often report carry-over effects without consistent primary task performance declines, and prior state manipulations (e.g., food deprivation) can alter within-subject variability in risk preferences. These strands motivate testing both mean shifts and stability (test–retest) of preferences and strategies under fatigue.

Design: Between-subjects manipulation with fatigue versus control, pre–post testing of economic tasks in gains and losses domains. Participants: 76 university students (41 males), ages 19–26 (M=22.3, SD=1.74). Initial sample: 44 randomized to fatigue (N=19) or control (N=25); later sample: additional 32 in an extended fatigue condition. Four fatigue-group participants were excluded for failure to report successful fatigue induction. Final analyzed sample: control N=25 (11 males; M=21.9, SD=1.44), fatigue N=47 (28 males; M=22.5, SD=1.87). Participants abstained from alcohol/caffeine for 24 hours. Procedure: Sessions (~2.5 hours) included consent, initial questionnaires (demographics, health, self-reported cognitive fatigue, Rating Scale of Mental Effort, RSME), pre-manipulation economic tasks (gains and losses), manipulation phase (fatigue vs control), post-manipulation questionnaires (fatigue and RSME), and repeat economic tasks. Payment: $10 base plus $0–10 bonus based on resolution of one randomly selected trial from each task instance (pre and post), resolved at end of session; no feedback provided during tasks. Manipulation: Fatigue group completed a 2-back task for 5 or 7 blocks of 300 trials (total 1500 or 2100 trials) over ~60–90 minutes. Each trial: 500 ms letter, 1.5 s response/feedback, 200 ms ITI. Targets were repeats of the letter from two trials earlier; right button for targets, left for non-targets. Implemented in MATLAB with Psychtoolbox. Control group watched relaxing videos (BBC The Life of Birds; Disney’s Beauty and the Beast; Disney’s Tangled) for ~90 minutes in a neutral wakeful state. Subjective measures: RSME (0–150; higher indicates more effort) and a single-item 1–10 self-reported cognitive fatigue question administered at baseline, after pre-task, after each N-back block (fatigue group) or matched intervals (control), and post-manipulation. Economic decision-making tasks: Two computerized tasks—gains and losses. Each trial required a choice between a certain option and a gamble (risky or ambiguous). Tasks were self-paced; one trial from each task instance was randomly selected for payment. Gains task: 135 risky and 30 ambiguous trials. Certain option $3–$7. Risky gambles had known winning probabilities (25%, 50%, 75%) and nine relative expected values (EV_gamble / V_certain: 0.5, 1.0, 1.3, 1.6, 1.9, 2.2, 2.5, 3.0, 3.5), with potential winnings $2–$98. Ambiguity trials had undetermined probabilities (randomly selected 0–1 prior to resolution), analyzed using expected value computed with 50% probability; rEV levels 0.5, 1.0, 2.0, 3.0, 4.0, 6.0; potential winnings $2–$168. Losses task: Mirror of gains with negative values. Ten rEV levels for risky and ambiguous trials: 0.1, 0.3, 0.5, 0.8, 1.0, 1.3, 1.5, 2.0, 3.0, 4.0. 150 risky and 50 ambiguous trials; potential losses $0.40–$112. Uncertainty preference metrics: For risk and ambiguity in each domain (gains, losses), preference quantified as a premium value. Choice functions plotted percent uncertain choices versus EV_gamble / V_certain (EVG/VC). Indifference point (first crossing at 50%) minus 1 yielded the premium. In gains: premium > 0 indicates risk/ambiguity aversion; < 0 indicates seeking; 0 neutrality. In losses: sign reversed (positive indicates seeking; negative aversion). Participants whose choice functions never crossed 50% were excluded for the specific analysis. Pre-manipulation preferences were also calculated using a power function; these correlated strongly with premium metrics (e.g., Gains risk r(48) = -0.497, p < .001; Losses risk r(66) = -0.640, p < .0001; Gains ambiguity r(38) = -0.773, p < .0001; Losses ambiguity r(62) = -0.877, p < .0001). Choice strategy metric: For each participant and domain, independent linear regressions related trial-by-trial choices to rEV (maximizing information) and to pWIN (satisficing information). Because rEV and pWIN were orthogonal across trials, their regressions captured distinct variance components. The choice strategy metric was the difference in r-squared values (rEV r^2 − pWIN r^2): positive values indicate maximizing, negative indicate satisficing, zero indicates equal reliance. Statistical analysis: SPSS v20 and MATLAB. Two-tailed tests, alpha p < .05. Primary comparisons examined change scores (post − pre) between fatigue and control via independent samples t-tests. Response times (median per participant) were analyzed similarly. Manipulation checks used t-tests and repeated-measures ANOVA with Greenhouse–Geisser correction across time points. Post-hoc, test–retest stability (pre vs post within subject) was compared between groups using Fisher r-to-z for correlation differences. Multiple comparisons were Bonferroni-corrected where specified. One outlier in gains risk difference (>5 SD) was excluded in gains analyses. Four participants in fatigue group who did not report successful fatigue induction were excluded from analyses.

Manipulation check: Baseline RSME and self-reported fatigue did not differ between groups. Across the session, fatigue participants reported significantly higher increases in effort and fatigue than controls (RSME change: t(65) = 8.78, p < .0001; cognitive fatigue change: t(69) = 7.57, p < .0001). Repeated-measures ANOVAs showed significant group-by-time differences for RSME (F(2.85,182.32) = 44.38, p < .0001) and self-reported fatigue (F(2.36,162.78) = 37.89, p < .0001) with no initial differences and significant differences at subsequent time points. N-back performance was maintained across blocks (misses: t(48) = −0.382, p = .704; false alarms: t(47) = −1.263, p = .213; correct % trend: t(48) = 1.840, p = .072). Response times: No effect of fatigue on change in response times in gains (Control M_diff = −0.48 s, SD = 0.48; Fatigue M_diff = −0.49 s, SD = 0.41; t(70) = 0.110, p = .913) or losses (Control M_diff = −0.52 s, SD = 0.40; Fatigue M_diff = −0.53 s, SD = 0.40; t(69) = 0.112, p = .911). Uncertainty preferences (mean shifts): No significant group differences in change scores for gains risk premium (Control M_diff = −0.12, SD = 0.21; Fatigue M_diff = −0.25, SD = 0.72; t ≈ 0.64, p ≈ .54), gains ambiguity premium (Control M_diff = −0.21, SD = 0.74; Fatigue M_diff = −0.42, SD = 1.28; t(38) = 0.531, p = .599), losses risk premium (Control M_diff = 0.04, SD = 0.40; Fatigue M_diff = −0.09, SD = 0.58; t(64) = 0.915, p = .364), and losses ambiguity premium (Control M_diff = −0.02, SD = 0.30; Fatigue M_diff = −0.01, SD = 0.55; t(62) = 0.080, p = .937). Choice strategies (mean shifts): No significant group differences in change scores for choice strategy in gains (Control M_diff = −0.10, SD = 0.19; Fatigue M_diff = −0.10, SD = 0.21; t(64) = 0.006, p = .995) or losses (Control M_diff = −0.01, SD = 0.14; Fatigue M_diff = −0.03, SD = 0.15; t(68) = 0.562, p = .576). No significant effects on component regressions in gains (rEV r^2: Control M_diff = −0.03, SD = 0.11; Fatigue M_diff = −0.06, SD = 0.09; t(67) = 1.338, p = .186; pWIN r^2: Control M_diff = 0.07, SD = 0.12; Fatigue M_diff = −0.03, SD = 0.14; t(64) = 1.048, p = .299) or losses (rEV r^2: Control M_diff = −0.002, SD = 0.104; Fatigue M_diff = −0.02, SD = 0.12; t(68) = 0.460, p = .647; pWIN r^2: Control M_diff = 0.003, SD = 0.05; Fatigue M_diff = 0.01, SD = 0.05; t(68) = 0.594, p = .555). Test–retest stability: Fatigue reduced stability (pre–post correlations) of several measures relative to controls. Risk premiums: gains z = 3.10, p = .002 (Control r = .94, p < .0001; Fatigue r = .56, p < .001); losses z = 2.24, p = .025 (Control r = .74, p < .0001; Fatigue r = .32, p = .039). Ambiguity premiums: no significant differences (gains z = 0.86, p = .389; losses z = 1.13, p = .259). Choice strategy: reduced stability in fatigue for gains (z = 2.07, p = .039; Control r = .89; Fatigue r = .68) and losses (z = 2.29, p = .022; Control r = .77; Fatigue r = .41). Components: losses pWIN r^2 stability reduced in fatigue (z = 2.42, p = .015); other component differences were nonsignificant. Correlations with subjective changes: Changes in subjective fatigue/effort did not significantly correlate (after Bonferroni correction) with changes in economic metrics. Within fatigue group, changes in subjective fatigue/effort did not correlate with N-back performance changes (after correction).

The study tested whether cognitive fatigue alters economic decision making by shifting uncertainty preferences or informational strategies. Contrary to the hypothesis of increased satisficing under fatigue, mean levels of risk/ambiguity preferences and choice strategies remained unchanged across gains and losses. However, cognitive fatigue significantly reduced test–retest stability of risk preferences and choice strategies, indicating that fatigue destabilizes decision making by increasing intra-individual variability. This instability implies reduced decision quality, as preferences and strategies become inconsistent across time, potentially violating rationality principles such as transitivity and increasing the likelihood of regret when revisiting choices in a rested state. The absence of mean shifts alongside reduced stability suggests that fatigue affects regulatory or consistency mechanisms rather than altering baseline preferences or strategy selection. The pattern resembles effects seen in ventromedial prefrontal cortex damage, which impairs preference consistency without slowing responses, hinting at overlapping functional mechanisms. Ambiguity preferences did not show reduced stability, potentially due to inherently lower reliability limiting power to detect fatigue effects. Despite increased subjective fatigue and effort, N-back performance was maintained, consistent with literature showing that primary task performance decrements are not always observed under fatigue while secondary measures are affected. Overall, cognitive fatigue compromises the consistency of economic choices without altering average preferences or strategies.

Cognitive fatigue, induced by prolonged demanding cognitive activity, does not shift average uncertainty preferences or the balance between maximizing and satisficing strategies. Instead, it destabilizes decision making by reducing the test–retest stability of risk preferences and choice strategies in both gains and losses. This increased intra-individual variability can undermine decision quality and rational consistency. Future research should identify the cognitive mechanisms (e.g., working memory, application of preferences, comparison processes) driving instability, examine longer or varied fatigue durations, and compare fatigue with other state alterations (e.g., sleep deprivation, ego depletion, aging) to map shared and distinct effects on economic decision processes.

• Ambiguity preference measures exhibited weaker test–retest reliability in controls, reducing power to detect fatigue-related stability changes in ambiguity.
• A subset of participants’ choice functions did not cross 50%, precluding preference estimation and reducing sample sizes for some analyses.
• The fatigue manipulation (60–90 minutes 2-back) increased subjective fatigue and effort but did not degrade N-back performance, raising questions about which primary tasks show time-on-task effects.
• The study used a between-subjects design with university students, which may limit generalizability.
• The manipulation duration was modest and may not capture effects of more prolonged or chronic fatigue; no differences were observed between 5- and 7-block fatigue groups.