What is choice architecture?
Thaler and Sunstein argue that no layout of a choice is neutral. The design claim is safe from any effect size, and the effect sizes are in dispute.
Answer to the question
What is choice architecture?
Choice architecture is the way options are laid out for a decision: their order, the default, the wording and numbers used, and the steps involved. Richard Thaler and Cass Sunstein named it in Nudge in 2008 and argue that no layout is neutral. The evidence is strongest for defaults, while the overall size of nudge effects remains disputed.
Key takeaways
- Thaler and Sunstein (2008; final edition 2021) argue that every form, menu or cafeteria has a layout, so the open question is whether it is designed well.
- Mertens and colleagues (2022) found in a PNAS meta-analysis that interventions changing the structure of a choice outperformed those changing the information or reinforcing intentions.
- Carroll and colleagues (2009) found that requiring new hires to make an explicit enrolment choice raised initial enrolment by 28 percentage points against a standard opt-in procedure.
- Camilleri and Larrick (2014) found preference for the efficient vehicle was highest when fuel economy was shown as the cost of gas over 100,000 miles.
- DellaVigna and Linos (2022) found an average effect of 1.4 percentage points across 126 nudge-unit trials covering 23 million people, against 8.7 in the academic literature, with selective publication explaining about 70 per cent of the gap.
In a study by Gabriel Carroll and colleagues, new hires were required to make an explicit choice about retirement enrolment, and the share enrolling at the start rose well above that under a standard opt-in procedure. The difference lay in how the decision was presented. That is choice architecture: the way a decision is laid out, from the order of the options and which one is already selected to how a number is expressed and how many steps lie between a person and the option they want. Richard Thaler and Cass Sunstein gave the field its name in Nudge, and their central claim is that every layout counts, including the ones nobody planned.
Who Arranges the Form?
Thaler and Sunstein make the case with a school cafeteria. Someone has to decide where the food goes, and whatever order that person picks will put some dishes at the front of the line and others at the back. Healthy items first is one design, healthy items last is another, and even a random order is a decision. They call the person who makes it a choice architect, and the term names a job that many people hold without having thought of it as one: the manager of a form, a menu, a website or a pension plan.
The authors draw a practical conclusion. Because some arrangement is unavoidable, the open question is whether it is designed well. In the final edition of Nudge they go further and say that opposing nudges as such cannot be a coherent position, since every system needs a rule for what happens when a person does nothing. Their preferred designs keep every option available and steer without forbidding.
The same edition names the dark side of the craft. Sludge, in their account, is friction that makes it harder for people to obtain an outcome that would leave them better off by their own judgement. A default whose opt-out is hard to use counts as sludge by this test, not as a nudge, and the authors recommend that organisations audit for it. The subject has its own piece in The Sludge That Keeps You Subscribed.
This makes the central claim of choice architecture unlike most findings in behavioural science. It is an argument about design, and no effect size can refute it. How strongly a particular layout moves people is an empirical question, and the evidence for it is uneven.
What Different Layouts Do
The most useful sorting of the evidence comes from a 2022 meta-analysis in PNAS by Stephanie Mertens and colleagues, which pooled choice architecture studies from many behavioural domains. Interventions that change how the choice is organised, such as altering a default, outperformed those that describe the alternatives differently or reinforce a person's existing intentions. Food choices were the most responsive category, and the paper reported a moderate publication bias in the literature it analysed.
The structure family has the longest record, and organ donation supplied its best-known comparison. In a 2003 policy forum in Science, Eric Johnson and Daniel Goldstein combined national data with experiments and reported large differences between countries that ask people to opt in and countries that ask them to opt out. They presented the gap as evidence that a default can shape expressed preference, with significant consequences. Countries differ in more than their consent rule, so a national comparison cannot isolate the rule by itself.
Inside a firm, Brigitte Madrian and Dennis Shea followed one large company that moved its retirement plan to automatic enrolment. Participation was significantly higher under the new arrangement. A substantial fraction of the employees hired under it kept both the default contribution rate and the default fund, even though few earlier hires had chosen that combination. The authors attribute this to inertia and to employees reading the default as investment advice. The case is also treated in Why You Never Change the Setting.
When the Form Asks Instead
The study that opens this piece shows the same inertia handled from the other side. Its authors, Carroll and colleagues including Madrian, report that required active choice raised initial enrolment by 28 percentage points against a standard opt-in procedure. They also built a model suggesting that active decisions suit settings where people tend to put decisions off and their preferences differ widely, while automatic enrolment suits settings where people lack the knowledge to judge the choice. The designer's job is to match the layout to the setting, and no single layout is best everywhere.
The information family acts on numbers. In two online experiments on how fuel economy is shown, Adrian Camilleri and Richard Larrick found that preference for the efficient vehicle was highest when economy was expressed as the cost of gas over 100,000 miles, whether or not the buyer's higher purchase price was offset by gas savings. The vehicle and the facts were the same and only the scale changed, a case taken further in Why Do MPG Figures Mislead Car Buyers.
| Case | Layout change | Reported result |
|---|---|---|
| Johnson and Goldstein, 2003 | Opt-out in place of opt-in for organ donation | Large differences between countries |
| Madrian and Shea, 2001 | Automatic enrolment in a retirement plan | Higher participation, many kept the default |
| Carroll and colleagues, 2009 | Required active choice | Initial enrolment up 28 percentage points |
| Camilleri and Larrick, 2014 | Fuel economy as cost over 100,000 miles | Highest preference for the efficient vehicle |
How Large Are the Effects?
Two letters in the same journal read the Mertens database differently. Maximilian Maier and colleagues reanalysed it with methods that model which results get published, and found no evidence for an overall effect, with evidence against an effect for the information and assistance categories. For the structure category, where defaults sit, their result was undecided rather than negative. Barnabas Szaszi and colleagues applied three bias-correcting methods to the same data and obtained average effects from slightly below zero to very small, with large variation between studies.
A separate test avoids the journals altogether. Stefano DellaVigna and Elizabeth Linos obtained every trial run by two large United States nudge units: 126 randomised trials covering 23 million people. The average effect on take-up was 1.4 percentage points, against 8.7 for nudge studies drawn from two meta-analyses of academic journals. The authors attribute about 70 per cent of the gap to selective publication, made worse by low statistical power, and most of the remainder to differences in the interventions themselves.
Our view is that the dispute and the definition do not collide, because they answer different questions. Thaler and Sunstein are themselves candid about scope: they write that light-touch design cannot solve every problem, and on climate change they argue that prices and regulation matter more than any rearrangement of options. These numbers suggest a modest average effect, so anyone planning a redesign should run a small trial of their own before expecting a published figure to repeat.
Common questions
What is choice architecture?
Choice architecture is the way options are laid out for a decision: their order, the default, the wording and numbers used, and the steps involved. Richard Thaler and Cass Sunstein named it in Nudge in 2008 and argue that no layout is neutral. The evidence is strongest for defaults, while the overall size of nudge effects remains disputed.
Who coined the term choice architecture?
Thaler and Sunstein introduced it in their book Nudge, published in 2008. Their earlier 2003 paper on libertarian paternalism set out the philosophy without using the term. The person who arranges a choice is the choice architect, whether or not they have ever thought of the work in those terms.
Is choice architecture the same as nudging?
Not quite. A nudge is an intervention that steers behaviour while keeping every option open, and choice architecture is the wider fact that every choice has a layout. Nudging is deliberate design within that fact. Sludge is the same craft aimed at making an outcome harder to reach.
Sources
Thaler, R. H., & Sunstein, C. R. Nudge. Seed book, first published 2008; the 2021 final edition is the one reported. Vault hub Books/Nudge.
Mertens, S., Herberz, M., Hahnel, U. J. J., & Brosch, T. (2022). “The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioral domains.” PNAS, 119(1), e2107346118. doi:10.1073/pnas.2107346118 https://doi.org/10.1073/pnas.2107346118
Maier, M., Bartos, F., Stanley, T. D., Shanks, D. R., Harris, A. J. L., & Wagenmakers, E.-J. (2022). “No evidence for nudging after adjusting for publication bias.” PNAS, 119(31), e2200300119. doi:10.1073/pnas.2200300119 https://doi.org/10.1073/pnas.2200300119
Szaszi, B., Higney, A., Charlton, A., Gelman, A., Ziano, I., Aczel, B., Goldstein, D. G., Yeager, D. S., & Tipton, E. (2022). “No reason to expect large and consistent effects of nudge interventions.” PNAS, 119(31), e2200732119. doi:10.1073/pnas.2200732119 https://doi.org/10.1073/pnas.2200732119
DellaVigna, S., & Linos, E. (2022). “RCTs to Scale: Comprehensive Evidence From Two Nudge Units.” Econometrica, 90(1), 81-116. doi:10.3982/ECTA18709 https://doi.org/10.3982/ECTA18709
Johnson, E. J., & Goldstein, D. (2003). “Do defaults save lives?” Science, 302(5649), 1338-1339. doi:10.1126/science.1091721 https://doi.org/10.1126/science.1091721
Madrian, B. C., & Shea, D. F. (2001). “The power of suggestion: Inertia in 401(k) participation and savings behavior.” Quarterly Journal of Economics, 116(4), 1149-1187. doi:10.1162/003355301753265543 https://doi.org/10.1162/003355301753265543
Carroll, G. D., Choi, J. J., Laibson, D., Madrian, B. C., & Metrick, A. (2009). “Optimal defaults and active decisions.” Quarterly Journal of Economics, 124(4), 1639-1674. doi:10.1162/qjec.2009.124.4.1639 https://doi.org/10.1162/qjec.2009.124.4.1639
Camilleri, A. R., & Larrick, R. P. (2014). “Metric and scale design as choice architecture tools.” Journal of Public Policy & Marketing, 33(1), 108-125. doi:10.1509/jppm.12.151 https://doi.org/10.1509/jppm.12.151