A Click Too Far: Separating Inattention from Avoidance in Food Choice

Last registered on September 28, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
A Click Too Far: Separating Inattention from Avoidance in Food Choice
RCT ID
AEARCTR-0019792
Initial registration date
September 21, 2026

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
September 28, 2026, 8:48 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Name
Cristiana Firullo
Affiliation
Cornell University
Contact Primary Investigator

Other Primary Investigator(s)

Additional Trial Information

Status
In development
Start date
2026-09-28
End date
2027-05-03
Keywords
Behavior, Education, Health, Welfare
Additional Keywords
nutritional disclosure, online grocery, rational inattention, information avoidance, food labeling, browser extension, field experiment
JEL code(s)
D83, D91, I12, L81, M31
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
Why does freely available information go unused? Two answers dominate the literature and are hard to tell apart. Attending to information is costly, or people would rather not know. They imply the same purchases and opposite policies: if attention is the constraint, simplifying the signal or making it more prominent helps; if people are turning away from it, those same interventions do nothing, and the case for acting on the product rather than on the label is stronger. Purchase data alone cannot distinguish them, because both produce the same basket. Food is the sharpest case. Whether a packaged product is healthy depends on its nutritional content, its additives and how processed it is. All of it is printed on the package, and none of it is easy to put together at the moment of choice. Online, it sits behind a click, on a page most shoppers never open. This study runs a randomised experiment inside real online grocery shopping. Adults in the United States who already order from Walmart.com at least weekly install a browser extension built for the study and complete one ordinary order, in their own account and with their own money. The extension varies at random, between participants, both how much nutritional information about each product is available to them and what it takes to see it. Two contrasts follow. Holding the effort fixed and varying the content asks what the detail behind a summary signal is worth. Holding the content fixed and varying the effort asks what it costs to look at information that is identical and one click away. Because the study observes not only what was bought but whether the participant engaged with the information at all, the two explanations come apart. If attending is what is costly, shoppers consult the score when the stakes are high, whatever they expect it to say. If they are avoiding it, they turn away precisely when they expect bad news about something they meant to buy. Engagement is observed on two margins: as an absence, where information offered is left unopened, and as an act, where information already on screen is closed and the number that was on screen is known. The primary outcome is the nutritional quality of the completed basket, measured as the average score of the items purchased, weighted by what was spent on each. A closing question with real money attached measures what continued access to the information is worth to participants. The extension runs only on Walmart.com. It records which products were displayed and which were purchased, under a randomly generated study code, never collects names, addresses or payment details, and stops collecting when the shopping session ends. Approved by the Cornell University Institutional Review Board, protocol IRB0151136.
External Link(s)

Registration Citation

Citation
Firullo, Cristiana. 2026. "A Click Too Far: Separating Inattention from Avoidance in Food Choice." AEA RCT Registry. September 28. https://doi.org/10.1257/rct.19792-1.0
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
Participants install a browser extension built for this study and complete one ordinary grocery order on Walmart.com, in their own account and with their own money. The extension changes how nutritional information about each grocery product is made available to them while they shop. Which version they receive varies between participants and is assigned at random. Participants are told in advance that the display varies and that it is decided by chance; they are not told which version they received until the study is over. The extension runs only on Walmart.com, records data under a randomly generated study code, never collects names, addresses or payment details, stops collecting when the shopping session ends, and is removed at the end of the study.
Intervention Start Date
2026-09-28
Intervention End Date
2027-05-03

Primary Outcomes

Primary Outcomes (end points)
Spend-weighted mean nutritional score of the grocery basket purchased during the session.
Primary Outcomes (explanation)
For participant i the outcome is Y_i = sum_j (s_j × e_ij) / sum_j e_ij, where e_ij is the dollars participant i spent on product j in the completed order and s_j is product j's nutritional score on a 0-100 scale, taken from the study's fixed reference table. Weighting by expenditure rather than by item count means the measure reflects what was actually bought rather than how many distinct products were tried.

The reference table is fixed and versioned before the first participant enters and is identical across arms and participants; scores are never recomputed at runtime. Items with no entry in the table are excluded from both numerator and denominator, and the share of basket expenditure so excluded is reported and tested for differences across arms. Baskets containing no scorable grocery item are excluded from the primary analysis and counted in the attrition table.

Three contrasts are pre-specified. Control versus T1 is the effect of the bare signal. T1 versus T2 is what the detail behind the signal is worth. T2 versus T3 is what it costs to look at information that is one click away and otherwise identical.

Secondary Outcomes

Secondary Outcomes (end points)
1. Share of basket expenditure on products scoring 60 or above.
2. Total basket expenditure, and number of distinct grocery items purchased.
3. Participant engagement with the nutritional information shown: whether it was opened or removed, how long it was on screen, and how that varies with the score and with whether the product was purchased.
4. Willingness to accept, elicited with real money, for keeping the extension installed until the next grocery purchase.
5. Products displayed with a score, considered, and not purchased.
Secondary Outcomes (explanation)
1. The threshold of 60 is fixed in advance at the boundary between the two middle bands of the score, computed on the same expenditure weights as the primary outcome. Robustness to thresholds of 50 and 70 is reported.
2. This is the mechanism test and the reason the study is run as an experiment rather than on purchase data. Two explanations for unused information predict the same baskets: that attending to it is costly, and that people prefer not to know. They separate in engagement rather than in outcomes, and the study observes engagement on two margins. Where the information is offered but not shown, disengagement is an absence: the placeholder is left unopened. Where it is shown without being asked for, disengagement is an act: it is closed, and the score that was on screen when it was closed is known. If attending is what is costly, engagement rises with the stakes attached to the product and does not depend on what the participant expects the score to say, which cannot be known before opening. If the information is being avoided, disengagement concentrates on products the participant intends to buy and whose score is unfavourable.
3. The participant states the minimum amount v they would accept to keep the extension installed until their next grocery checkout. A price p is drawn from an announced distribution, independent of v. If p is at least v the participant keeps the extension and is paid p; otherwise the study ends there. Payment is conditional on the extension still being installed at the next purchase, and participants are told this before stating v. Keeping the extension bundles continuing to receive the information and continuing to be observed. In the control arm no information is shown, so v measures the cost of being observed alone; in the treated arms it measures that cost net of the value of the information. The difference between arms is the revealed dollar value of the disclosure, and the ordering across T1, T2 and T3 prices the detail separately from the bare signal. The control arm is not an ordinary comparison group here: it is what makes the other three interpretable.

Experimental Design

Experimental Design
Individually randomised four-arm experiment, conducted in participants' own browsers during a single real grocery purchase on Walmart.com. Equal allocation. Randomisation occurs at the end of the enrollment questionnaire, after all baseline characteristics have been recorded and before any study content is shown.
Experimental Design Details
Not available
Randomization Method
Computer-generated, by the randomiser built into the Qualtrics survey software, at the end of the enrollment questionnaire. Equal allocation across the four arms, no stratification. The assignment is written to the participant's survey record as an embedded field and read from there by the browser extension, so the arm recorded in the survey and the arm delivered on Walmart.com cannot diverge.
Randomization Unit
Individual participant. There is a single level of randomisation; no group or session-level assignment.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
2,000
Sample size: planned number of observations
2,000 completed grocery orders, one per participant. At least 2,000 former grocery orders (one per participant).
Sample size (or number of clusters) by treatment arms
500 participants control, 500 participants treatment 1, 500 participants treatment 2, 500 participants treatment 3.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Primary outcome, spend-weighted mean basket score on a 0-100 scale. Assuming a participant-level standard deviation of 14 points, 500 per arm gives 80 percent power at a two-sided 5 percent level, Holm-adjusted across the three pre-specified contrasts, to detect a difference of 0.21 standard deviations, or 2.9 score points. Under 20 percent attrition to 400 per arm this becomes 0.23 standard deviations, or 3.2 points. Secondary outcome, share of basket expenditure on products scoring 60 or above, treated as a continuous participant-level variable with a standard deviation of 0.25: 5.1 percentage points at 500 per arm, 5.7 at 400. The standard deviation is assumed, not observed. It will be re-estimated on the pilot and this calculation updated before the main data collection, with the change logged.
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
IRB0151136
IRB Approval Date
2026-09-18
IRB Approval Number
N/A
Analysis Plan