---
name: Status quo bias
slug: status-quo-bias
category: Decision making
categories: [Onboarding, Pricing, Retention]
evidence: moderate
page: /principles/status-quo-bias/
updated: 2026-10-15
---

# Status quo bias

> People leave things as they are — the plan, the setting, the provider they already have — even when changing would cost them nothing.

## What it is

An option carries extra weight because it is the one already in place: switching costs effort and uncertainty, the current arrangement reads as the recommended one, and the person being asked to change it is rarely the one who set it up.

## Evidence

Samuelson and Zeckhauser (1988) named the effect with a questionnaire. 486 students answered hypothetical problems — a budget split, an inherited portfolio, a job at another college, the colour of an ordered wagon — in versions where the options were listed neutrally and versions where one was described as what the respondent already held. An option was picked most often in the status quo position, less often under neutral framing, and least often as the alternative to someone else's status quo. Pooled, their fitted summary was SQ = .17 + .83 × NEUT: occupying the status quo position added roughly 17 percentage points to an option's share, and in relative terms the biggest beneficiaries were the alternatives nobody otherwise wanted. The origin is thinner than its citation count suggests — hypothetical scenarios, no money, student subjects, and the status-quo-versus-alternative difference reached the 10% significance level in only 31 of 54 comparisons — so the 17 points are a direction, not a figure to size a screen with.

The field half of that paper is the part worth building on. Of 9,185 Harvard employees choosing among eight health plans in 1986, only about 3% switched plan in a year. Because new hires chose under neutral framing, they work as a control group: the incumbent Blue Cross plan took between 6.4% and 24.7% of new enrollees across four age bands, against 27.4% to 50.0% of those enrolled before 1980 — two to four times the share, for the same plan in the same year. The retirement data is more striking and less conclusive. TIAA-CREF allocations barely moved between 1981 and 1986 despite large swings in the two funds' returns, only 28% of surveyed participants had ever changed their split and fewer than 2.5% changed in a given year, even though changing was free and almost all of them knew it. But new and old participants chose similarly, which the authors say plainly: those differences were “for the most part not great in percentage terms”. What a long-tenured user has set tells you less about their preference than what a new user picks on the same screen.

The experiments have been replicated, and the failure is the instructive part. Xiao, Lam, Piara and Feldman (2021) ran two preregistered replications on Mechanical Turk (n = 311 and n = 316) of four of the original scenarios and found strong support in three — budget allocation, portfolios, the job move — with odds ratios mostly between 1.7 and 4.9 against the non-status-quo framing. The wagon colour scenario produced nothing: no significant difference in the expected direction in either phase, and three comparisons pointing the other way. Their explanation is a design spec. In that scenario the status quo had been assigned by the dealer rather than chosen, it had existed for two days, the respondent did not yet own the wagon, and changing it cost “a word to the wagon dealer”. The status-quo-versus-neutral contrast — the one a product actually faces — was the inconsistent one in both the original and the replication.

Where real money is involved the inertia is large and well documented. Madrian and Shea (2001) studied a firm that switched to automatic 401(k) enrollment: participation was 86% among employees hired under it, against 37% for the cohort hired just before at the same tenure. The revealing figure is what people did next. 61% of the automatic-enrollment cohort sat at the complete default — enrolled, contributing 3%, 100% in the money market fund — against 1% of every other cohort, so the pattern is inertia rather than preferences that happened to match. Handel (2013) found the same inertia in health plan choices and then the uncomfortable consequence: policies that reduced it improved individual choices but worsened adverse selection enough to double the existing welfare loss. Hortaçsu, Madanizadeh and Puller (2017) found Texas households rarely searched for an alternative electricity retailer and attached a brand advantage to the incumbent, with low-cost information the lever that moved surplus.

Two results tell you what happens when you deliberately break inertia. Carroll, Choi, Laibson, Madrian and Metrick (2009) made 401(k) enrollment an unavoidable question for new hires: initial participation rose 28 percentage points over standard opt-in, producing in three months a savings distribution that standard enrollment took three years to reach. The advantage decayed — 17 points at 24 months, 5 points at 42 — so forcing a decision mostly moves it forward in time. Ascarza, Iyengar and Schleicher (2016) ran the opposite experiment by accident: a telecoms operator proactively recommended cheaper, better-fitting plans to customers, and churn over the next three months rose from 6% in the control group to 10% among those contacted, which they read as the campaign lowering inertia and making past usage salient. Reopening a settled decision reopens all of it, including the decision to stay.

The organ-donation literature is where this principle is most often oversold. Johnson and Goldstein's (2003) online experiment is smaller than its reputation: 161 respondents, with consent at 42% under opt-in, 82% under opt-out and 79% with no default at all. Opt-out did not differ significantly from the no-default condition — only opt-in was significantly lower, so the finding is a penalty for making people act, not a windfall from presuming consent. Their country comparison separates four opt-in from six opt-out registries by nearly 60 percentage points of effective consent, and the Netherlands mailed more than 12 million letters to a population of 15.8 million without shifting its rate. Consent is not transplants, though. Shepherd, O'Carroll and Ferguson (2014) found across 48 countries that opt-out ones had higher deceased donor rates (14.24 against 9.98 per million) and more kidneys transplanted (28.32 against 22.43) but fewer living donors (5.49 against 9.36). Arshad, Anderson and Sharif (2019), using 2016 data for 35 OECD countries, found no significant difference in deceased donors (20.3 against 15.4) or total solid organ transplants (63.6 against 61.7), only fewer living donors (4.8 against 15.7). McLaughlin and Mays (2024), reporting the official evaluation of England's deemed consent law, state that consent rates have not increased since it took effect in May 2020, attributing that to a soft opt-out bolted onto the existing opt-in system and families still acting as gatekeepers. The screen-level claim survives; the systems claim does not.

Meta-analysis puts a number on the screen-level claim. Jachimowicz, Duncan, Weber and Johnson (2019) pooled 58 default studies at d = 0.68, with wide variation including non-significant results and two reversals, and defaults strongest where they read as an endorsement or as the status quo. The wider nudge literature has had a harder decade: Mertens and colleagues (2022) report d = 0.45 across more than 200 studies and note moderate publication bias; Maier and colleagues (2022) re-analysed the same data and found the overall effect indistinguishable from zero — but the “structure” category that contains defaults was the one case left undecided (0.12 to 0.23), and Szaszi and colleagues (2022) named defaults as the explicit exception to their conclusion that large consistent nudge effects should not be expected. In real interfaces the asymmetry is enormous: Utz and colleagues (2019) report Cookiebot data covering 3 million consent decisions in which 98.84% accepted with all boxes pre-ticked and 5.59% with none, and Nouwens and colleagues (2020) found 93.1% of consent interactions never left the first page. Hence the grade: robust in direction, replicated, and huge where a user is asked to leave a stated arrangement alone — but the famous applied numbers inflate it, the hypothetical origin barely reaches significance, and it disappears when the user did not choose the arrangement and can change it for free.

## When it holds

- The arrangement has to be the user's own. The one scenario the preregistered replication failed was the one where the status quo was assigned by someone else, was two days old, was not yet owned and could be changed for free.
- Switching has to cost something — effort, uncertainty or a judgement the user does not feel qualified to make. The scenarios that replicated all left room to infer a transition cost; the one that did not spelled out that change was costless.
- It is strongest against actively choosing an alternative, weakest against a clean slate. Johnson and Goldstein's opt-out condition did not differ from having no default; only the opt-in condition was lower.
- Defaults read as recommendations work harder. Effects were strongest where the shipped state conveyed a recommendation or reflected the status quo, and larger in consumer than environmental domains.
- It is largest where the user never had a preference. Only one in three TIAA-CREF participants called their initial allocation an informed choice and one in four called it a guess; almost half took the simple 50/50 split.
- Asking again dissolves it. Proactively recommending a better-fitting plan raised three-month churn from 6% to 10%, because the campaign lowered inertia and made usage salient.
- Forcing an active choice substitutes for it only briefly. Compelled enrollment decisions beat opt-in by 28 percentage points initially, 17 at 24 months and 5 at 42.
- Long tenure is not evidence of fit. Harvard's incumbent health plan held two to four times the share among pre-1980 enrollees as among new hires choosing in the same year, and Texas households attached a brand advantage to the incumbent retailer they had never searched away from.
- Nothing is known about repeated exposure in software. Every figure here comes from one-off decisions — a plan election, a consent dialog, a questionnaire — not from a setting a user sees weekly.

## How to apply

- On any migration or plan change, put the user's current state on the screen next to what would replace it, field by field, and name what carries over. That transition cost is the lever is an inference from the replication's failed scenario and Hortaçsu's low-cost-information result; the screens themselves are untested.
- Carry state across the change rather than asking for it again: import the data, keep the existing names, preserve the layout, and show the import as complete before asking for anything new. Same inference, same lack of direct evidence.
- Set the shipped state to the one the user would pick for themselves, not the one that collects most. For data settings that is a legal requirement, not a preference: GDPR Article 25(2) requires that by default only personal data necessary for each specific purpose are processed.
- Make leaving cost the same as arriving. Count the steps from signed-in to cancelled against the steps from landing page to paying, and make the shorter one the target — DSA Article 25(3)(c) names a termination procedure harder than subscription as a practice the Commission may issue guidelines on.
- Where the decision matters and the user has no preference yet, ask once as an unavoidable active choice at first run, and expect the advantage to decay: 28 percentage points initially, 5 by 42 months.
- Before any prompt that reopens a settled plan or setting, size the churn risk. The one field experiment that recommended better plans to existing customers raised churn from 6% to 10%.
- On a redesign or forced migration, keep the previous arrangement reachable as a named, dated option rather than a hidden flag, and measure how many users return to it. Treat that as a test you are running, not a finding you have.
- Show the user what they already have at the moment you ask them to change it — the configured workspace, the saved history, the plan they are on. Defaults that read as the status quo were the strongest cell in the meta-analysis.

## What to avoid

- Pre-selecting the data-hungry option and relying on nobody changing it. The EDPB calls this Deceptive Snugness, names the default effect as its mechanism, and places it in conflict with GDPR Article 25(2); the Cookiebot figures show what it buys — 98.84% acceptance pre-ticked against 5.59% not.
- Burying the reject, downgrade or cancel path one layer deeper than the accept path. Removing the reject-all button from the first page of a consent dialog raised acceptance by 22 to 23 percentage points, and 93.1% of interactions never reached a second page at all.
- Re-asking a question the user has already answered until they answer it differently. DSA Article 25(3)(b) singles out repeatedly requesting a choice already made, especially through pop-ups that interfere with the experience.
- Running renewals on the expectation that the user has forgotten. The EU-wide 2023 CPC sweep of 399 online shops found 148 using at least one dark pattern, 23 of them hiding information specifically to push consumers into a subscription.
- Treating the US click-to-cancel rule as the compliance floor. The Eighth Circuit vacated the FTC's 2024 Negative Option Rule in Custom Commc'ns, Inc. v. FTC, and the FTC recodified the pre-2024 text effective 12 February 2026 — the EU prohibitions and state auto-renewal laws did not go away with it.
- Quoting organ donation as proof that defaults deliver outcomes. Consent rates differ by nearly 60 percentage points between registry types while deceased donor rates showed no significant difference across 35 OECD countries, and England's opt-out law has not raised consent rates.
- Reading retention as satisfaction. 61% of one automatic-enrollment cohort sat at a 3% contribution in a money market fund against 1% of everyone else, which is a measure of how little the arrangement was ever chosen.

## Measuring it

- Primary metric: the change rate on the setting, plan or provider in question, split by whether the user configured the current state themselves or inherited it from you. These are two populations and only the second is status quo bias.
- Secondary: drop-off on the migration path — import started against import finished — and the share of users still on the shipped state after 30, 90 and 180 days, against a cohort asked to choose actively. Transition cost is the part of the mechanism a build can move; a shipped-state share near 100% means the interface chose, not the user.
- Backfire metric: cancellation, downgrade and churn in the 90 days after any screen that reopens a settled decision, against a holdout that was never prompted. The one field test of this put the gap at 6% against 10%.
- Backfire metric: reversals and regret signals — users switching a setting back, support tickets about a change they did not expect, refund requests on renewals. Inertia that was holding unwanted state shows up here, not in retention.
- Backfire metric, for consent and privacy screens: the acceptance rate gap between your shipped state and an unticked version of the same screen. A gap the size of the Cookiebot asymmetry is evidence the interface, not the user, is deciding.

## Live examples

- Booking.com puts the stay the guest already holds beside the one they asked for, striking through Thu, Mar 19, 2026 and Fri, Mar 20, 2026 above Mon, Mar 23, 2026 and Tue, Mar 24, 2026, and setting an original price of $55.74 against a new price of $55.74, so what the change moves and what it leaves alone are both on the screen before it is confirmed.
- Spotify badges the plan the listener is already on as “Current” and spells out its terms — 1 Premium account, 15 hours a month of audiobook listening, cancel anytime — directly above Duo at US$16.99 / month with 2 Premium accounts, so the arrangement being left is described in the same words as the one that would replace it.
- Monzo names what travels before asking for the switch — salary, direct debits and payments moved in one go under the Current Account Switch Guarantee — and offers a second route that only updates the employer, so the work of leaving the old account is stated rather than left to be guessed at.
- Amazon Shopping prints the state each permission is currently in next to its name — camera enabled for 27 features, bluetooth, contacts, local network, location, microphone, motion and photos all disabled — so a change begins from what is already set rather than from a bare request.
- HBO Max offers “Accept All” and “Manage Cookies” as the only two buttons on the first screen of its cookie dialog, so accepting every purpose listed takes one tap while refusing any of them takes a further step into a second screen.

## References

- Samuelson & Zeckhauser (1988). Status Quo Bias in Decision Making. Journal of Risk and Uncertainty. — https://doi.org/10.1007/BF00055564
- Xiao, Lam, Piara & Feldman (2021). Revisiting Status Quo Bias: Replication of Samuelson and Zeckhauser (1988). Meta-Psychology. — https://doi.org/10.15626/MP.2020.2470
- Madrian & Shea (2001). The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior. Quarterly Journal of Economics. — https://doi.org/10.1162/003355301753265543
- Carroll, Choi, Laibson, Madrian & Metrick (2009). Optimal Defaults and Active Decisions. Quarterly Journal of Economics. — https://doi.org/10.1162/qjec.2009.124.4.1639
- Handel (2013). Adverse Selection and Inertia in Health Insurance Markets: When Nudging Hurts. American Economic Review. — https://doi.org/10.1257/aer.103.7.2643
- Hortaçsu, Madanizadeh & Puller (2017). Power to Choose? An Analysis of Consumer Inertia in the Residential Electricity Market. American Economic Journal: Economic Policy. — https://doi.org/10.1257/pol.20150235
- Ascarza, Iyengar & Schleicher (2016). The Perils of Proactive Churn Prevention Using Plan Recommendations: Evidence from a Field Experiment. Journal of Marketing Research. — https://doi.org/10.1509/jmr.13.0483
- Johnson & Goldstein (2003). Do Defaults Save Lives? Science. — https://doi.org/10.1126/science.1091721
- Shepherd, O'Carroll & Ferguson (2014). An International Comparison of Deceased and Living Organ Donation/Transplant Rates in Opt-in and Opt-out Systems: A Panel Study. BMC Medicine. — https://doi.org/10.1186/s12916-014-0131-4
- Arshad, Anderson & Sharif (2019). Comparison of Organ Donation and Transplantation Rates between Opt-out and Opt-in Systems. Kidney International. — https://doi.org/10.1016/j.kint.2019.01.036
- McLaughlin & Mays (2024). What Does the Evaluation of the Organ Donation (Deemed Consent) Act 2019 in England Tell Us About the Effectiveness of Deemed Consent Systems for Deceased Organ Donation? Transplantation. — https://doi.org/10.1097/TP.0000000000005246
- Jachimowicz, Duncan, Weber & Johnson (2019). When and Why Defaults Influence Decisions: A Meta-analysis of Default Effects. Behavioural Public Policy. — https://doi.org/10.1017/bpp.2018.43
- Mertens, Herberz, Hahnel & Brosch (2022). The Effectiveness of Nudging: A Meta-analysis of Choice Architecture Interventions across Behavioral Domains. PNAS. — https://doi.org/10.1073/pnas.2107346118
- Maier, Bartoš, Stanley, Shanks, Harris & Wagenmakers (2022). No Evidence for Nudging after Adjusting for Publication Bias. PNAS. — https://doi.org/10.1073/pnas.2200300119
- Szaszi, Higney, Charlton, Gelman, Ziano, Aczel, Goldstein, Yeager & Tipton (2022). No Reason to Expect Large and Consistent Effects of Nudge Interventions. PNAS. — https://doi.org/10.1073/pnas.2200732119
- Utz, Degeling, Fahl, Schaub & Holz (2019). (Un)informed Consent: Studying GDPR Consent Notices in the Field. ACM SIGSAC Conference on Computer and Communications Security. — https://doi.org/10.1145/3319535.3354212
- Nouwens, Liccardi, Veale, Karger & Kagal (2020). Dark Patterns after the GDPR: Scraping Consent Pop-ups and Demonstrating their Influence. CHI Conference on Human Factors in Computing Systems. — https://doi.org/10.1145/3313831.3376321
- Regulation (EU) 2016/679 (GDPR), Article 25(2): by default, only personal data necessary for each specific purpose are processed. — https://www.legislation.gov.uk/eur/2016/679/article/25
- European Data Protection Board (2023). Guidelines 03/2022 on Deceptive Design Patterns in Social Media Platform Interfaces, version 2.0: Deceptive Snugness. — https://www.edpb.europa.eu/system/files/documents/2023-02/edpb_03-2022_guidelines_on_deceptive_design_patterns_in_social_media_platform_interfaces_v2_en_0.pdf
- Regulation (EU) 2022/2065 (Digital Services Act), Article 25: online interface design and organisation. — https://publications.europa.eu/resource/celex/32022R2065
- European Commission (2023). Consumer Protection: Manipulative Online Practices Found on 148 out of 399 Online Shops Screened. CPC Network sweep, IP/23/418. — https://ec.europa.eu/commission/presscorner/api/files/document/print/en/ip_23_418/IP_23_418_EN.pdf
- Federal Trade Commission (2026). Revision of the Negative Option Rule to Conform to Federal Court Decisions. Federal Register. — https://www.federalregister.gov/documents/2026/02/12/2026-02866/revision-of-the-negative-option-rule-withdrawal-of-the-cars-rule-removal-of-the-non-compete-rule-to

## Similar principles

- Default effect — /agents/default-effect.md
- Loss aversion — /agents/loss-aversion.md
- Choice overload — /agents/choice-overload.md
- Door-in-the-face — /agents/door-in-the-face.md
- Self-determination theory — /agents/self-determination-theory.md