An energizing role for motivation in information-seeking during the early phase of the COVID-19 pandemic

The paper addresses whether human information-seeking is irrational or can be understood as goal-rational through the lens of utility and motivation. Against views that curiosity is a goal-independent drive leading to non-utility-maximizing behavior, the authors hypothesize that a latent motivational state explains when and why people seek information. The COVID-19 pandemic provided a natural context to test this: domain relevance varied widely across individuals, allowing measurement of motivation (COVID-19 concern), expected utility (usefulness judgments), and epistemic behavior (seeking, satisfaction, memory). The core predictions were: (1) a directing effect—higher domain-specific motivation should increase seeking of information in that domain via increased expected utility; and (2) an energizing effect—motivation should raise the average utility of information, increasing seeking even for unrelated content. They further predicted these effects would be dissociable from non-specific anxiety.

The authors situate their work within longstanding claims that human information-seeking is often non-instrumental or irrational, influenced by curiosity as a goal-independent drive (e.g., Loewenstein, Berlyne, Kidd & Hayden). They review evidence that curiosity states enhance learning and share neural correlates with reward processing, suggesting overlaps with goal-directed mechanisms. They note rational analyses of curiosity emphasizing incentive value and novelty/complexity, and learning-based theories of motivation that map states to action values. They contrast accounts that must exempt trivia from utility considerations with mechanistic theories positing that all behavior, including trivia seeking, is utility-based. Prior work on reinforcement learning and prediction errors informs their hypothesis that satisfaction versus expectation (prediction error) should shape subsequent information-seeking.

Design and participants: 6,135 US-based participants were recruited on Amazon Mechanical Turk with twice-weekly waves between March 11 and May 7, 2020; after exclusions, 5,376 participants (median age 36; range 18–89; 2,818 female) were analyzed. Session 2 (recall) occurred 7–8 days later; 71.48% of eligible participants returned. IRB approval obtained; informed consent collected. Stimuli: 104 questions total. COVID-19-related (n=52) sourced from WHO, CDC, NYT; half designed to be clearly useful and half conventionally non-useful. General questions (n=52): half trivia (from prior studies) and half useful household tips. An exploratory crowdsourced general category appeared last and was not analyzed. Tasks and measures: - Information-seeking (waiting task): On each trial, participants saw a question and chose: “know” (if they knew the answer), “wait Xs” (4, 8, 12, or 16 s randomly assigned) to receive the answer, or “skip”. Task duration was fixed at 300 s, independent of choices. The proportion of wait vs skip across delays indexed information-seeking. After answers, participants rated satisfaction (1–5 Likert). - Expected utility: After the waiting task, participants rated a held-out set of five questions per type (COVID-19-related and general) on expected usefulness for self and others (1–7 Likert), without seeing answers. Different items were used to avoid response repetition biases. - Memory: 7–8 days later, participants attempted to recall answers for previously waited-for items; free-text responses were coded for accuracy by a blind rater. - Motivational states: End of session 1 questionnaires assessed COVID-19 concern (domain-specific motivation; items spanned medical, economic, social concerns, severity and risk perception) and non-specific anxiety/affect (control). Measures were validated via Bayesian PCA (three components; COVID-19-specific, negative affect, positive affect), correlations with real-world events (job loss, income decrease, self-isolation), social distancing indices, and temporal trends. Exclusions: Removed participants for technical issues (n=4), <perfect English fluency (n=358; 5.84%), excessive app switching during tasks (n=335; 5.46%), >20% non-responses (n=4), or mean RT >2 SD faster than group (n=58). For session 2, excluded >5 app interactions (n=176; 3.27%) or non-compliant recall responses (n=52; 0.97%). Analyses: Multilevel logistic regression for trial-level waiting and memory; multilevel ordered-logistic regression for satisfaction and usefulness. Predictors included COVID-19 concern, non-specific anxiety (control), question type (COVID-19-related vs general), judged usefulness (item-level), and wait duration. Usefulness ratings were modeled with an ordinal Item Response Model to obtain metric values. Mediation analysis used a joint multilevel model to test whether usefulness judgments mediated the effects of COVID-19 concern and its interaction with question type on waiting. A separate model assessed prediction-error effects on subsequent-trial waiting (satisfaction and usefulness on prior trial predicting next-trial waiting). Models used maximal random effects, Hamiltonian Monte Carlo (Stan), regularizing priors, four chains, convergence diagnostics reported. Analyses conducted in R 3.6.0, Stan 2.23.0, Julia 1.4.2.

- Directing effect of motivation: Higher COVID-19 concern increased waiting for COVID-19-related questions relative to general questions (interaction b=0.11, 95% PI [0.08, 0.14]). COVID-19 concern also increased judged usefulness for COVID-19-related versus general questions (interaction b=0.10, 95% PI [0.07, 0.12]). Mediation showed a significant indirect effect of the interaction on waiting via usefulness (b=0.02, 95% PI [0.002, 0.04]); 22.27% of the effect mediated (95% PI [1.71%, 49.86%]). - Energizing effect of motivation: COVID-19 concern increased waiting for general (non-COVID) questions (simple effect b=0.17, 95% PI [0.09, 0.26]) and elevated overall usefulness judgments (main effect b=0.51, 95% PI [0.47, 0.55]). Indirect main effect on waiting via usefulness was significant (b=0.14, 95% PI [0.01, 0.25]); 69.27% mediated (95% PI [5.38%, 151.16%]). - Satisfaction: COVID-19 concern predicted greater satisfaction with all answers (main effect b=0.43, 95% PI [0.38, 0.48]), especially for COVID-19-related answers (interaction b=0.08, 95% PI [0.06, 0.11]). - Learning dynamics: Prediction errors influenced subsequent information-seeking. Satisfaction on trial t predicted more waiting on trial t+1 (b=0.10, 95% PI [0.05, 0.15]); usefulness expectation on trial t predicted less waiting on t+1 (b=-0.09, 95% PI [-0.13, -0.05]). The difference (satisfaction − usefulness) corresponds to a positive prediction-error effect increasing subsequent seeking. - Memory: COVID-19 concern was associated with poorer memory for general information, whereas memory for COVID-19-related information was spared (interaction pattern as in Fig. 3c). Non-specific anxiety was not significantly associated with memory (Fig. 3d). - Dissociation from non-specific anxiety: Non-specific anxiety negatively predicted waiting for all questions (main effect b=-0.28, 95% PI [-0.37, -0.20]) and lowered usefulness judgments (b=-0.10, 95% PI [-0.15, -0.04]), especially for COVID-19-related questions. COVID-19 concern and non-specific anxiety were moderately correlated (r=0.44) but had opposite associations with epistemic behavior. - Measure validation: COVID-19 concern tracked real-world factors (higher with job loss, income decrease, self-isolation; increased mid-March 2020; higher in states with more social distancing).

Findings support a dual role for motivation in epistemic behavior. Domain-specific motivation (COVID-19 concern) directed information-seeking toward relevant content by increasing expected utility for that content, consistent with goal-rational behavior. Motivation also energized information-seeking more broadly by elevating the average utility of pursuing information, increasing seeking even for unrelated questions and enhancing satisfaction. Trial-by-trial reinforcement-learning-like signals (prediction errors) shaped subsequent choices, indicating participants maintained an average value signal for information-seeking. These results challenge accounts requiring a special, non-instrumental curiosity drive and instead align information-seeking with general reward-based computational mechanisms, wherein information and reward share common value codes. The dissociation from non-specific anxiety underscores the specificity of motivational influences and provides insight for public health: targeting motivational states that elevate perceived utility could steer and boost constructive information engagement during crises.

The study demonstrates that motivation both directs and energizes human information-seeking by modulating expected utility, reconciling apparent irrationality with goal-rational behavior. By measuring domain-specific concern, utility expectations, and epistemic behavior in a large, naturalistic sample during early COVID-19, the authors show that motivated states increase seeking and satisfaction for relevant information and broadly elevate information pursuit, including for unrelated content. Reinforcement-learning dynamics further explain how experienced utility updates future behavior. These insights offer a framework for predicting and shaping information-seeking in public policy and education. Future work should causally manipulate motivational states in controlled settings, disentangle highly motivating versus demotivating influences, generalize beyond pandemic contexts, and integrate models of expected informational gain with utility-based accounts.

The design is largely correlational and naturalistic, limiting causal inference about motivation’s effects. The COVID-19 concern measure was unidimensional, potentially conflating motivating and demotivating facets. The unique context of the early pandemic may limit generalizability to typical conditions. While expected utility was modeled, the study did not explicitly quantify expected information gain or participants’ prior knowledge dependencies. Some exploratory stimuli categories were omitted from analysis, and online data collection may introduce uncontrolled environmental variability.