Designing plant-based defaults to maximize uptake while minimizing reactance

Last registered on July 06, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
Designing plant-based defaults to maximize uptake while minimizing reactance
RCT ID
AEARCTR-0019102
Initial registration date
July 03, 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
July 06, 2026, 9:31 AM EDT

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

Locations

Country
United States of America
Region
online

Primary Investigator

Name
Seth Green
Affiliation
Stanford University
Contact Primary Investigator

Other Primary Investigator(s)

PI Name
Maya Mathur
PI Affiliation
Stanford University
Contact Investigator
PI Name
Kenjin Chang
PI Affiliation
Greener By Default
Contact Investigator
PI Name
Natalia Lawrence
PI Affiliation
University of Exeter
Contact Investigator

Additional Trial Information

Status
In development
Start date
2026-07-13
End date
2026-08-24
Keywords
Behavior
Additional Keywords
nudge, behavioral science, plant-based defaults, defaults
JEL code(s)
C91, D91, D12, Q18
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
Plant-based defaults are widely touted as a scalable, sustainable solution for increasing uptake of plant-based foods. However, defaults are not a singular thing; their designs can be varied to make opting out more or less difficult, or they can be accompanied by persuasive text that promotes the change to choice architecture to participants' conscious awareness. This experiment -- a collaboration between researchers at Stanford, the University of Exeter, and the nonprofit Greener By Default -- compares four plant-based defaults in an online context on two dimensions: how much they increase uptake of plant-based food and how much reactance they engender (re: how much they annoy people). We will test four plant-based defaults in the setting of a hypothetical catered lunch to see which performs best on these two axes: a standard default where the plant-based option is pre-selected and a participant can switch to the meat option, a nonstandard default where the plant-based option is pre-selected and a user must unclick it (and then select nothing) to get a meat-based lunch, a standard plant-based default along with a message about inclusivity, and a standard plant-based default along with a message about sustainability. We will measure reactance by asking how satisfied participants were with their meal order. We will also conduct some exploratory analyses based on demographic information collected.
External Link(s)

Registration Citation

Citation
Chang, Kenjin et al. 2026. "Designing plant-based defaults to maximize uptake while minimizing reactance." AEA RCT Registry. July 06. https://doi.org/10.1257/rct.19102-1.0
Sponsors & Partners

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

Request Information
Experimental Details

Interventions

Intervention(s)
Participants imagine attending an event with a free catered lunch and choose between two meals: a Tofu Bánh Mì (plant-based) and a Chicken Bánh Mì. We randomize the format in which this choice is presented. In the control condition, neither option is pre-selected. Four treatments make the plant-based meal the default: T1 pre-selects it on a standard two-option ballot; T2 presents a single pre-checked checkbox that participants must uncheck to receive the chicken option instead; T3 and T4 use the T1 ballot plus a short justification, framing the plant-based meal as the most inclusive option (T3) or the most sustainable option (T4).
Intervention (Hidden)
All arms open with a common vignette: participants imagine they are going to
an event with a free, catered lunch and are told the caterers want to know how
many of each meal to provide. Control and T1 then ask participants to select
which of two options they would prefer, with no default in control and the
Tofu Bánh Mì pre-selected in T1. T2 shows one pre-checked box for the Tofu
Bánh Mì and instructs participants to uncheck it if they wish to have the
Chicken Bánh Mì instead. T3 adds to the pre-selected ballot: "We have found
this to be the most inclusive option that meets the most diners' needs,
including religious groups and people with allergies." T4 instead adds: "We
have found this to be the most sustainable option, with significantly lower
greenhouse gas emissions than meat-based meals." The two sandwiches are
described with identical ingredient lists except the protein. All participants
then complete the same endline items: a manipulation check (which meal did
you order), satisfaction with the meal chosen, satisfaction with the available
choices, satisfaction with how the choices were presented, an estimate of the
percentage of participants choosing the same meal, and an open comment box.
The survey takes 1-2 minutes.
Intervention Start Date
2026-07-13
Intervention End Date
2026-08-24

Primary Outcomes

Primary Outcomes (end points)
(1) Choice of the plant-based meal (chose_veg). (2) Dissatisfaction with the chosen meal (dissat_chosen).
Primary Outcomes (explanation)
chose_veg = 1 if the participant ends up with the Tofu Bánh Mì and 0 if the
Chicken Bánh Mì. In the arms with a two-option ballot (control, T1, T3, T4),
this is the option selected. In T2 (opt-out checkbox), leaving the pre-checked
box checked codes as 1 and unchecking it codes as 0; a blank checkbox field
is a deliberate opt-out, not a missing value. dissat_chosen = 1 if the
participant rates satisfaction with the meal they chose (1 = extremely
dissatisfied to 5 = extremely satisfied) in the bottom two boxes (<= 2).
Confirmatory criteria: H1 holds for an arm if its estimated uptake gain over
control is significant at p < .01 (linear probability model, HC2 robust
standard errors) and meets a pre-registered minimum of 10 percentage points
(T1), 20 (T2), or 30 (T3, T4). H2 holds for an arm if its dissatisfaction rate
and the upper bound of the rate's 95% Wilson confidence interval both fall
below a pre-registered ceiling of 20% (T1, T3, T4) or 30% (T2). Control has no
registered ceiling; we report its rate against a 10% descriptive benchmark.

Secondary Outcomes

Secondary Outcomes (end points)
Dissatisfaction with the available meal choices; dissatisfaction with how the
choices were presented; manipulation check (reported meal matches recorded
choice); perceived descriptive norms (estimated percentage of participants
choosing the same meal). Pre-specified exploratory analyses: pooled framed
defaults (T3 + T4) vs. the bare default (T1) on both primary outcomes;
cross-condition comparisons among treatment arms; dietary identity coded from
open text; heterogeneous treatment effects by gender, education, and age.
Secondary Outcomes (explanation)
The two secondary satisfaction items use the same 1-5 scale and bottom-two-box
dissatisfaction coding as the primary reactance outcome; both are reported
descriptively with 95% Wilson confidence intervals and carry no registered
ceilings. The manipulation check is the share of participants whose reported
meal matches their recorded choice. Norms are measured on a 0-100 slider.
Dietary identity is coded from an optional free-text item, with coding
finalized after inspecting real responses. Exploratory analyses carry no
confirmatory status.

Experimental Design

Experimental Design
Five-arm between-subjects survey experiment on Prolific (US adults, N =
3,750). Participants imagine a catered event and choose between a plant-based
and a chicken sandwich under one of five randomly assigned choice formats: no
default, plant-based pre-selected, plant-based opt-out checkbox, pre-selected
plus an inclusivity message, or pre-selected plus an environmental message.
Each format is evaluated on two co-primary outcomes: plant-based uptake,
tested against a pre-registered minimum gain over control, and dissatisfaction
with the chosen meal, tested against a pre-registered acceptability ceiling.
Analyses are linear probability models with HC2 robust standard errors
(unadjusted primary, covariate-adjusted robustness) and 95% Wilson intervals
for dissatisfaction rates, with alpha = .01 for significance tests. The
analytic sample is completed responses passing a pre-treatment attention
check, the sole exclusion, applied before any analysis. A pre-analysis plan
containing the full analysis pipeline, run end-to-end on simulated data, is
attached.
Experimental Design Details
The attached pre-analysis plan specifies the complete pipeline: renaming from
the raw Qualtrics export, variable construction, exclusions, the H1/H2
decision criteria, power analyses, and secondary and exploratory analyses, all
run end-to-end on simulated data so the real export can be dropped in without
further analytic decisions. For unaddressed contingencies we defer to the
Lin, Green, and Coppock standard operating procedures, v1.05
Randomization Method
Computer randomization: a randomizer element in the Qualtrics survey flow
presents the five arms evenly; assignment is stored in an embedded-data field.
Randomization Unit
individual
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
3,750 individuals (individual-level randomization; no clustering).
Sample size: planned number of observations
3,750 US adults recruited on Prolific; approximately 3,560 expected in the analytic sample after attention-check exclusions.
Sample size (or number of clusters) by treatment arms
750 individuals per arm: control, T1 (pre-selected default), T2 (opt-out
checkbox), T3 (inclusivity framing), T4 (environmental framing); approximately
712 per arm in the analytic sample.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Primary behavioral outcome (chose_veg, a proportion): with approximately 712 analytic observations per arm, alpha = .01 two-sided, and an assumed control uptake of 20%, power is 96.4% to detect a 10 percentage point gain (the smallest registered threshold) and effectively 100% for the 20 and 30 percentage point thresholds. Primary reactance outcome (dissat_chosen): by simulation (10,000 draws per arm), the probability that the 95% Wilson upper bound falls below the registered ceiling is 93-94% for T1, T3, and T4 (assumed true rate 15%, ceiling 20%) and 99.7% for T2 (assumed true rate 22%, ceiling 30%). The H1 criterion also requires the point estimate to meet the threshold, so the design is powered to certify effects comfortably above the thresholds, not effects exactly at them.
Supporting Documents and Materials

Documents

Document Name
Claude wrote a registered report, why not?
Document Type
other
Document Description
totally AI generated registered report because it's fun to see what Claude comes up with
File
Claude wrote a registered report, why not?

MD5: dbbab030551cbb6432106b6fa1a4498c

SHA1: 7b4a4a0bd6fdafe08b59095a35036ba5fbf44082

Uploaded At: July 03, 2026

Document Name
Survey
Document Type
survey_instrument
Document Description
File
Survey

MD5: 70c873bfeb1ba1dfd746ca31ad8e8901

SHA1: ffca405777d3a5a345bfdc8bac4a3df7a9dc0403

Uploaded At: July 03, 2026

IRB

Institutional Review Boards (IRBs)

IRB Name
Stanford University Administrative Panels for the Protection of Human Subjects (IRB)
IRB Approval Date
2026-04-08
IRB Approval Number
85867
Analysis Plan

Analysis Plan Documents

GBD pre-analysis plan

MD5: 4f2c5461a1127f8ab08f46a2375ea6a3

SHA1: f5afa549db506604ac314f9e2347b24c384bac32

Uploaded At: July 03, 2026

Post-Trial

Post Trial Information

Study Withdrawal

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

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials