Attention Allocation under Scarcity: Theory and Evidence

Participants are randomly assigned to different budget experiences in the consumption task before facing a common later decision environment. In the initial adaptation rounds, participants complete the consumption task with either large or small budgets, and a simple productivity task. In the subsequent test round, all participants face the same (medium) budget and incentives. To control for potential learning effects in the consumption task, there is no overlap on the solutions of the test round to the previous faced problems. The intervention is designed to study the effects of the perception of scarcity while keeping objective resources fixed, so the main comparison is between the two groups in the test round.

2026-06-22

2026-11-30

The main dependent variables will be measured in the common-budget test round:

1. Proportion of subjects perceiving scarcity in each group (self reported)
2. Attention allocated to each task, measured via mouselab.
3. Performance in the production task (cognitive sliders).
4. Decision quality in the consumption task.

Perceived scarcity is measured after each round using a question that asks participants whether they felt they had enough budget, on a 1–5 Likert scale. The main measure will be a binary indicator for reporting that the budget was not enough. We will also analyze the original response scale.

Attention allocated to the consumption task is measured using mouse-tracking data. The main measure is the share of total task time spent on the consumption task.

Performance in the production task is measured by the number of correctly completed sliders in the test round.

Decision quality in the consumption task is measured primarily as the points obtained from the selected bundle divided by the maximum points attainable under the participant’s budget in that round, regardless of whether the bundle satisfies the payment eligibility requirement. This puts performance on a comparable scale across budget conditions. We will also report the actual payoff earned in the consumption task, which equals zero when the participant does not select one item from each essential category.

For the main analysis, we will use all participants with valid outcome data in the test round. As a robustness analysis, we will repeat the main outcome analyses among participants who appear to have engaged with both tasks, defined as those who spent at least 5% of total task time in each task.

Secondary outcomes include total earnings across both tasks, the number of items revealed, the number of modifications in the selected bundle, whether the participant selected the optimal bundle, the number of sliders attempted, task-switching behavior, confidence about performance in the consumption task, and self-reported difficulty of each task. We will also report decision quality conditional on selecting essential items in the consumption task. Finally, we will collect self-reported income and financial stress to examine their relationship with perceived scarcity and the main treatment effects.

Total earnings are computed as the sum of earnings from both tasks. The number of items revealed measures how much of the consumption choice set the participant observed before choosing. The optimal-bundle indicator equals one if the participant selects the payoff-maximizing bundle under the given budget. Task-switching behavior is measured using transitions between the two tasks. Confidence and perceived difficulty are elicited after each round. Self-reported income and financial stress are elicited after the experiment is complete and will be used for descriptive and exploratory heterogeneity analyses.

This is an experiment conducted on Prolific. Participants complete an incentivized dual-task environment in which they divide their attention between a consumption task and a separate productivity task. The experiment first gives participants different budget histories over several adaptation rounds and later places everyone in the same test-round decision environment. The main comparison studies whether participants behave differently in this common environment, where actual resources and incentives are held fixed.

Not available

Randomization is done by a computer within the experimental software. Treatment assignment is randomized at the participant level and balanced across conditions as closely as possible.

The unit of randomization is the individual participant.

No

0

500 participants

250 participants will be assigned to the high budget adaptation condition (treatment) and 250 participants will be assigned to the low budget adaptation condition (control).

Sample size was determined based on the least-powered primary outcome in the pilot study. In the pilot, the full-sample difference in this outcome was approximately 0.046 on a 0-1 normalized outcome scale, with a pooled standard deviation of approximately 0.195. Since the theoretical predictions are directional, the main preregistered tests for the primary outcomes will use one-sided comparisons in the predicted direction. With a 5% significance level and 80% power, detecting the pilot effect for the least-powered primary outcome requires approximately 225 participants per treatment arm. We therefore plan to recruit 500 participants in total, corresponding to 250 participants per arm. At this sample size, the minimum detectable effect is approximately 0.043 units, or 4.3 percentage points, on the normalized outcome scale. Participants who drop out before completion or whose data cannot be used because of technical problems affecting the primary outcomes will be replaced. We will also report two-sided p-values and confidence intervals for transparency.