Choice blindness is a failure to notice that an outcome presented as your own choice is not the option you originally selected. Under some experimental conditions, people go further: when asked why they chose the substituted outcome, they provide coherent reasons for it. The lesson is not that people never know what they want. It is that a verbal explanation after the event is not always reliable evidence of what was chosen before the event.
The phenomenon was named in research published by Petter Johansson, Lars Hall, Sverker Sikström and Andreas Olsson in Science in October 2005. Their experiment used controlled substitutions between a person’s stated choice and the item later presented for explanation. Follow-up studies examined other domains, including written attitudes. This article explains the precise intention–outcome mismatch, how to calculate detection rates honestly and how to design fair confirmation systems without treating an experimental finding as a licence to deceive people.
Picture a student who selects one of two course schedules. At a later review, a document displays the other schedule as if it were the student’s own selection. If the student fails to notice and begins explaining why the displayed schedule seems sensible, the explanation may reflect the displayed information as much as the earlier private choice. The scene is invented to illustrate a possible mechanism, not a report of research on students or a claim that schools commonly swap decisions.
Contents: a practical route through the evidence
Follow the basic mechanism from original choice and the 2005 study to the four-stage sequence, mismatch monitoring, later reasons, and genuine changes of mind.
Check the science and arithmetic: related change blindness, post-event misinformation, moral attitude research, the political survey research case, event denominators, participant denominators, and weighted group comparisons.
Use it safely: confident reasons, interface design, organisational records, learning environments, surveys, ethics, limitations, independent transfer, decision checks, FAQs, primary research and next reading.
Start with the reader question that matters to you
If you want the core idea, focus first on the initial decision, the substituted outcome, the opportunity to detect a mismatch and the explanation that follows. Those four elements define why choice blindness is not identical to simply making a poor decision or forgetting an unrelated fact. The key comparison is between a traceable earlier choice and a later item represented as that choice.
If you are reading as a teacher or parent, begin with everyday decision records and how to invite checking without embarrassing the chooser. A learner may sincerely give a good-sounding reason for an option that was accidentally copied into the wrong place. The appropriate response is to verify the actual record, not accuse the learner of dishonesty.
If you are responsible for a form, workplace decision or digital interface, read the chapters on identity of the chosen option, confirmation screens, audit trails and independent checks. The relevant repair is a clearer link from selection to receipt, with enough detail for a user to recognise the option and correct errors. A confirmation box is meaningful only if it allows a realistic check.
If you are assessing the research, follow the material about experimental manipulation, event-level versus participant-level denominators, detection timing and inference boundaries. The phenomenon is important precisely because it can be measured against an independent record. But one task does not establish a universal rate of blindness across domains, populations or stakes.
Begin with the choice as it actually happened
A person sees alternatives and selects one. This initial choice may involve preferences, attention, available evidence and immediate context. The person might be certain, tentative or even indifferent. Choice blindness research does not require assuming that the person’s preferences are fully stable or perfectly articulated. It requires knowing which alternative the person indicated in the moment.
In a laboratory, that indication can be observable: pointing to a photograph, ticking a response or selecting a visible item. The investigator can then compare the recorded response against the outcome later displayed. Without that independent observation, an after-the-fact explanation may be interesting, but a mismatch between intention and outcome has not been established.
Consider a fictional restaurant exercise where a participant chooses meal A rather than meal B. The record says A. The server later presents B as the selected meal. That does not prove choice blindness by itself. The error becomes informative only if there is a fair opportunity to detect it and the researcher records whether the person notices, corrects the choice or accepts the substitution.
The distinction between intended option and later outcome is the anchor for honest interpretation. In real life a person can change their mind, receive new information or intentionally delegate a choice. Those events are not automatically errors. The same observed final option can result from a genuine change, an accidental substitution or manipulated feedback; evidence is needed to tell them apart.
What the 2005 experimental demonstration actually did
The original research asked participants to choose between pairs of pictured faces on the basis of attractiveness. On selected trials a controlled sleight-of-hand substitution caused the item presented afterwards to be the unchosen face. Participants were also asked to explain their choice. This allowed the researchers to compare a documented original selection with a different displayed outcome.
Some participants noticed the mismatch immediately or later. Others did not detect it and nevertheless offered reasons for the presented option. The official Lund University research record describes the finding as failure to detect mismatches between intention and outcome, alongside introspective reports supporting the substituted alternative. This is the research question, not a demonstration that all human reasoning is fabricated.
The 2005 paper appeared in Science, volume 310, pages 116–119, DOI 10.1126/science.1111709. The original sample and procedure are tied to that design. A lecture may retell the study as if every participant accepted every swap; that would erase the detection cases and convert an observed experimental pattern into a false generalisation.
A responsible reader asks who was studied, how many trials were swapped, when participants could notice, what counted as a detection and how explanations were evaluated. The distinction matters because the proportion of trials noticed and the proportion of people who notice at least one trial can be very different. Both can be true without contradicting each other.
The four-stage mechanism in order
Stage one is selection: an individual makes a discriminable choice between visible or otherwise defined alternatives. Stage two is substitution or misattribution: the visible feedback or item described as selected differs from the actual recorded choice. Stage three is monitoring: the individual has some chance to compare the new display with the earlier selection. Stage four is the response: the person may flag a mismatch, accept it, express uncertainty or explain the displayed result.
Choice blindness is observed when a real mismatch is not noticed within a specified opportunity. Post-hoc rationalisation is a further possible response when someone explains a substituted result as though it arose from their own earlier choice. The two should not be collapsed into one universal behaviour. A person may fail to notice a swap yet refuse to explain, or notice later upon a second look.
An explanation is not a machine-readable record of the exact psychological process that produced an earlier decision. People infer reasons from memory, present evidence, social context and their beliefs about themselves. When the item on display has changed, that input may shape what feels like a sensible explanation. The research tests a limitation on introspection; it does not prove that all reasons lack value.
Place the stages on a timeline before making claims. If a participant changes their preference after receiving new information, the study must distinguish that from an unnoticed substitution. If they notice a switch and deliberately accommodate it, the event is not undetected blindness. If the experiment hides relevant details or rushes a participant, detection opportunity is part of the causal setting.
Why mismatch monitoring can be harder than it sounds
Everyday attention is selective. Someone may focus on whether the displayed item has appealing properties rather than on comparing it feature by feature with a recently chosen alternative. Short retention intervals do not guarantee accurate identity matching, especially when two options are visually similar or the task does not signal that identity verification is important.
The brain does not preserve a perfect high-resolution record of every decision and its visual context. Memory for an option can emphasize a broad feature such as pleasant expression rather than every detail. A later alternative sharing those features may feel compatible with the earlier decision. This is one plausible route to a missed discrepancy, not a claim that a single brain process explains every choice-blindness result.
Social expectations may also matter. If a person assumes the experimenter or interface will present the option they selected, they may have little reason to perform an explicit comparison. Trust in the procedure shifts attention away from error detection. But the research does not establish that trust alone caused the original results; multiple task features can contribute.
For an interface, the generalisable lesson is modest but useful: never rely entirely on a user’s feeling of continuity. Make the identity of the selection inspectable, preserve its description and offer a correction path. That is a design response to possible human oversight, not an assertion that every mismatch will be missed.
What post-hoc reasons can and cannot tell us
When an experimenter asks why someone chose an outcome, the person can examine that outcome and generate a plausible story around its features. If the outcome has secretly been substituted, the reasons may still sound coherent. That is startling because listeners often assume a reason is a faithful record of the preceding decision process.
One possibility is reconstruction: the speaker combines current perception with general preferences and fills gaps in memory. Another possibility is an ordinary preference for consistency in conversation. Neither can be established for a particular person merely from the existence of a neat explanation. Experimental designs can compare groups or trials, but an individual account still has uncertainty.
Calling every such explanation a deliberate lie misunderstands the central insight. A participant who does not recognise the mismatch might be speaking sincerely from the evidence currently available to them. The finding concerns the limits of introspective access and monitoring, not a blanket moral judgment about the integrity of people who explain their choices.
Conversely, some explanations may correctly reflect earlier preferences even when an experimenter attempted a substitution, and some participants immediately challenge an incorrect presentation. A careful article retains those possibilities. The most defensible policy is to treat explanations as evidence of a person’s present interpretation, then compare them with the documented choice when accurate history is needed.
Choice blindness is not the same as genuinely changing your mind
A person who chooses A, later reconsiders and knowingly prefers B has changed their mind. If they acknowledge that shift, there is no hidden mismatch to detect. Choice blindness concerns an A-to-B substitution represented as continuity, not a transparent update in preference. This is why a time-stamped before-and-after record is more informative than asking people to declare that their current opinion is identical to their earlier one.
Suppose a committee member initially selects a training course for Tuesday but later discovers the participant works Tuesday mornings. Switching to a Thursday course is sensible adaptation. It should be recorded as a change with a reason, not classified as choice blindness. A good process celebrates legitimate revision while detecting clerical substitutions.
The same caution applies to research surveys. Responses to value-laden questions can change with framing, information and conversation. A changed response is not proof that a hidden substitution influenced the participant unless the protocol documents both the original answer and the altered feedback. Failing to preserve that distinction would mistake flexibility for failure.
Give a decision log two separate fields: what was originally selected, and what is now approved. If those differ, capture the reason or acknowledgement. This design allows users to revise a position without having to insist that they always held it. It also makes accidental swaps easier to identify.
How choice blindness differs from change blindness
Change blindness concerns a failure to notice a change in a visual scene or display, often when it occurs across an interruption. Choice blindness is narrower: the unnoticed change is presented as the outcome of the person’s own earlier choice. The personal decision and sometimes the generated explanation are what make the latter especially interesting.
A person might miss that a wall poster changed colour without ever having chosen either version; that illustrates change detection. In a choice-blindness experiment, the mismatch concerns the specific option the person selected, and the protocol may ask them to justify the displayed alternative. The two phenomena can share attention and memory constraints without being interchangeable.
To see the separate mechanism, read the live eduKate specialist Change Blindness — Scene Interruptions, Missed Changes and Reliable Visual Checks. Its reader job is to understand visual comparison failures. This article instead owns the selection-to-outcome link and the implications for accounts of personal decisions.
In an educational demonstration, label the two types separately. Show a changed classroom display to discuss visual attention. Show a recorded student selection and a different returned selection to discuss outcome verification, but do not covertly manipulate real students without appropriate ethical approval and protections. Learning about a cognitive effect is not permission to deceive classmates.
How choice blindness differs from post-event misinformation
Post-event misinformation concerns later information altering what someone recalls or reports about an earlier event. Choice blindness starts with a documented choice and an altered outcome presented as if it were the chosen one. The phenomena can intersect when the altered feedback shapes recollection, but the experiments answer different questions.
Imagine a witness who sees a bicycle and is later asked a leading question describing it as a motorcycle. That illustrates a potential misinformation pathway. A choice-blindness scenario instead has a person directly pick bicycle picture A and subsequently receives bicycle picture B under the label their selection. The point of comparison is the recorded selection.
The existing Post-Event Misinformation — Leading Questions, Source Confusion and Fair Recall Tests goes further into how memory reports can be affected by later suggestions. Readers should use that article for eyewitness-question and recall-specific jobs rather than treat all memory effects as equivalent.
An independent source record is valuable in both cases, but for different reasons. For choice blindness it anchors which option was actually selected; for post-event misinformation it helps separate what was witnessed from information introduced after the event. In neither case should an observer assume a mistaken report proves intentional deception.
What later research on moral attitudes showed
In 2012, Lars Hall, Petter Johansson and Thomas Strandberg reported a study in PLOS ONE using a paper survey of moral attitudes that could covertly present a reversal of selected earlier responses. Their aim was to test whether participants would detect those reversals and how they would reason when confronted with the altered statements.
The published abstract reports that 69 percent of participants failed to detect at least one of two manipulated changes. That is a participant-level measure under the study’s design, not a general statement that 69 percent of all opinions can be swapped undetected. Different counts and detection windows would lead to different percentages, even with similar behaviour.
The authors observed that people sometimes gave articulate reasons supporting the reversed view. The study broadened the phenomenon beyond a rapid aesthetic choice, while leaving important questions about individual differences, context, demand characteristics and the meaning of expressed attitudes. It did not prove that there are no sincere convictions.
Interpret this work as evidence that the relation between stated attitudes, immediate feedback and self-explanation deserves examination. It should not be used to dismiss somebody’s ethical beliefs, to label dissenters irrational, or to design unconsented manipulation in education, marketing or politics. The ethical boundary is part of applying cognitive science responsibly.
A political survey is a research case, not an operating manual
A further study published in PLOS ONE in April 2013 examined choice-blindness manipulations of responses to Swedish election-related survey issues. Participants received altered feedback in a controlled research setting, after which researchers investigated detection and reported intentions. It was a study of attitude measurement and self-attribution, not a general-purpose technique for influencing elections.
The published abstract reports a low observed detection proportion for manipulated replies and describes subsequent willingness among some participants to consider shifting coalition preference. Those findings are tied to the sample, setting, question wording and study period. They cannot establish that a particular present-day voter would change a vote, or that reported willingness translates into actual election-day behaviour.
Political questions are personally meaningful and socially contested. It is especially important to distinguish what participants said, what they noticed and how the researchers interpreted the evidence. A measurement about openness to reconsidering an attitude should not be presented as an argument for any party or a justification for covert influence.
For schools or workplace teams, the ethical takeaway is verification: survey results should be traceable to the responses participants actually gave, and corrections should be possible. Do not use the study to secretly rewrite opinions. A process that manufactures apparent consent damages the integrity of the very measurement it claims to improve.
Worked example: detection per swapped outcome
Consider an invented controlled research dataset with 24 participants, each exposed to three manipulated decisions. That produces 24 times three, or 72 manipulated trials. Suppose 12 manipulations were detected immediately and nine more were detected later, with those nine drawn from cases not already counted. The number of distinct detected manipulated trials is 12 + 9 = 21.
The immediate detection proportion is 12 divided by 72, or approximately 16.67 percent. Detection at any recorded time is 21 divided by 72, approximately 29.17 percent. The undetected count is 72 minus 21 = 51, so the undetected proportion is 51 divided by 72, approximately 70.83 percent. All figures are fabricated to teach denominators; they are not results from the original Science experiment.
The word distinct is important. If later interviews simply rediscover three of the same twelve immediately noticed trials, counting those again as new detections would double-count the same event. The correct any-time numerator is the number of unique manipulated trials detected by the end of the observation window, not the number of mentions or opportunities to notice.
The denominator is 72 manipulated trials, not all ordinary and manipulated trials combined. If each of the 24 participants also completed ten unmanipulated trials, that adds 240 opportunities for normal choices but does not make those 240 swaps available to detect. Mixing them into the denominator would artificially lower the apparent detection rate.
Why a participant detection rate cannot be inferred from trial counts
Continue the fictitious 24-person example with 21 detected manipulations. It would be wrong to say that 21 of 24 people noticed a mismatch. One person may detect more than one manipulated trial. Without knowing how those 21 detections are distributed across the participants, the participant-level numerator is unavailable.
At one extreme, if each person has at most three manipulated trials, seven participants might account for 21 detections by noticing all three each. At another extreme, 21 different participants might account for one detection each. Both yield 21 out of 72 event detections, but the percentage of people noticing at least one change would differ greatly: 7 divided by 24 is approximately 29.17 percent versus 21 divided by 24, or 87.5 percent.
The range is a mathematical illustration, not a confidence interval and not a claim about plausible human behaviour. It exposes the missing data structure. A real report should identify participants, manipulated trials per participant and whether detections happened during the task or later. When observations cluster within people, uncertainty calculations must also respect that clustering.
This is why the 2012 abstract’s participant-level wording cannot be compared directly with a study reporting the fraction of manipulated trials detected. Before combining percentages, ask exactly what counted as the unit in the numerator and denominator. Different denominators can tell complementary truths, but not without the corresponding data.
Weighted comparisons: do not average percentages blindly
Suppose two hypothetical conditions are evaluated. Group A has 20 manipulated trials and eight are detected, so its detection proportion is 8 divided by 20 = 40 percent. Group B has 50 manipulated trials and ten are detected, so its proportion is 10 divided by 50 = 20 percent. The groups differ in their trial counts.
An unweighted average of those percentages gives (40 + 20) divided by two = 30 percent. But the pooled trial detection proportion is (8 + 10) divided by (20 + 50) = 18 divided by 70, or approximately 25.71 percent. The difference is not an error in basic arithmetic; the two summaries answer different questions.
The equal-weight average asks what the mean rate would be if both group-specific rates were given equal summary weight. The pooled proportion asks what share of all 70 manipulated trials were detected. If a research question concerns participants rather than trials, even this pooled event calculation may not be the right estimator.
Before interpreting a reported difference, check group size, selection, number of manipulations per participant, detection definition and interval. Percentages on their own cannot establish why one group detected more. They may reflect visual difficulty, task conditions, different explanations, sampling variation or an experimental treatment; inference requires an appropriate design.
Why a confident explanation is not proof of an accurate memory
A fluent explanation can sound like a window into the speaker’s mind. It may contain concrete adjectives, consistent reasoning and personal references. Yet a person can produce such an account using the outcome presently displayed. The form of a statement and its correspondence to a historical choice are separate questions.
In the original face-choice context, a participant might explain the substituted face by referring to the expression or apparent friendliness of that displayed face. The explanation may be coherent as a description of the visible alternative while failing to describe why the unobserved original decision occurred. This is not a paradox if verbal reasoning is partly reconstructive.
A listener should not infer intentional fabrication solely from a mismatch. Nor should an investigator assume every post-choice reason is worthless. People can accurately report some preferences, motivations and constraints. The scientific challenge is knowing when a later account is an appropriate measure of an earlier decision and when a direct behavioural record is required.
Good review questions separate present evaluation from historical reconstruction. Ask what the speaker thinks of the displayed option now, then independently check what they selected earlier. That two-step design permits honest changes of mind and prevents the demand for consistency from suppressing useful corrections.
Build decision interfaces that make mismatches easy to catch
Show a decision summary immediately after selection, using the same distinctive label and meaningful details as the original option. A user who selected a Monday morning class should see Monday, time, campus and fee terms where relevant, not simply a generic statement that enrolment succeeded. The summary should be readable before any irreversible commitment.
For important transactions, produce a receipt that retains the chosen option, timestamp, reference number and version of the relevant terms. A receipt is not proof that every term was understood, but it creates an inspectable record. Where edits are allowed, record changes with previous and new values so that genuine revision is visible instead of silently overwriting history.
Do not make confirmation a test of whether the user can recite an internal product code. A sensible interface presents natural language alongside identifiers and permits correction without penalty. A student selecting a programme should be able to compare the meaningful programme name and session, and a worker confirming a roster should see the actual shift rather than an ambiguous abbreviation.
Design checks in proportion to stakes. A mistaken colour preference may need an easy undo; a payment instruction or access-permission change needs stronger independent verification. The purpose is not to burden every selection with bureaucracy, but to remove the assumption that a human will reliably notice a substituted outcome without assistance.
Choice blindness in group and organisational records
Groups generate multiple possible points where an outcome can drift from what members intended: notes, draft minutes, summaries, circulated approvals and final execution records. A board might discuss option A but later receive a summary that sounds like option B. This does not mean the board is exhibiting choice blindness as an experimentally verified phenomenon; it is a risk to be controlled with provenance.
Preserve an explicit decision register that distinguishes proposed alternatives, the final selected option, its rationale, dissent and any later amendment. Read critical details aloud at approval time when appropriate, then issue a traceable record. If a summary differs, members should have a straightforward opportunity to correct it before action follows.
An assistant-generated meeting summary can make this risk sharper when it condenses discussion into polished prose. The summary might express a plausible recommendation that nobody actually approved. A human verifier should compare consequential commitments against the source minutes and recorded vote or owner approval rather than equating linguistic fluency with authenticity.
Organisations can also suffer from group decision biases such as privileging shared information. That is a different reader job from individual outcome substitution. The distinction helps leaders choose repairs: broaden evidence sharing when relevant facts were overlooked, or strengthen record and approval integrity when an outcome is attributed to the wrong decision.
Teaching choice blindness without humiliating learners
Students are often asked to explain answers after receiving marked work. A learner may sincerely explain the answer currently highlighted, especially if the original working has been separated from the submitted record. Rather than accuse the learner of inconsistency, a teacher can request the original steps and compare them with the marked output.
Teach a three-column reflection: my original choice, the feedback I received and what I believe after reviewing the evidence. The columns keep change of mind visible and make error correction legitimate. A student who changes a mathematical answer after discovering a mistake should be able to document that improvement without being labelled unreliable.
In vocabulary or reading work, a learner may select one meaning in a multiple-choice question and later receive a worksheet with a different option circled. The teaching question is whether they can check the evidence in the text and their original mark. The cognitive phenomenon does not substitute for subject knowledge; it motivates a practical verification habit.
Classroom demonstrations should be transparent or conducted under appropriate ethical arrangements. Secretly swapping a student’s meaningful course choices or grades is not a justified teaching exercise. Safer examples use clearly fictional records, intentionally labelled errors and permission to correct outcomes without embarrassment.
What survey researchers should protect
A survey response often passes through forms, exports, databases, charts and reports. A mismatch can arise from a reversed code, a mistaken row join or an edited question label even without experimental sleight of hand. Choice-blindness research is one reason not to assume respondents will identify all discrepancies when confronted with an altered summary.
Give each question a stable identifier and preserve its version and response coding. In a test dataset, if 1 means agree on one scale but disagree on a reverse-coded item, that transformation must be explicit. Validate mappings independently and check displayed answer text against original raw values before publishing group summaries.
Do not interpret a participant’s apparent acceptance of a false display as consent to change the recorded response. Ethical survey work needs accurate reporting, appropriate participant protections, and documented correction procedures. Experimental deception is subject to review and informed ethical constraints, not something ordinary organisations may borrow for persuasion.
For analysis, distinguish a raw response, an adjusted scale score and a researcher’s interpretation. The survey studies cited here used manipulations in controlled experiments; ordinary survey operations should do the opposite, preserving true answers and providing a clear audit path when material errors are discovered.
Why the research does not authorise manipulation
The fact that a substitution can go unnoticed is a warning about vulnerability, not a user-manual for exploiting it. Deliberately changing another person’s recorded choices can violate trust, research ethics, professional duties or law depending on context. Organisations should use the evidence to improve transparency, not to manufacture apparent approval.
Any experimental use of deception or concealed manipulation needs formal consideration of consent, participant welfare, necessity, debriefing and independent ethics review where applicable. The original researchers worked in defined study settings. Copying a trick into a workplace, school, medical form or political campaign without safeguards does not reproduce legitimate research conditions.
The practical tools in this article are protective: clear descriptions, independently verifiable receipts, correction routes, versioned documents and accurate denominators. These reduce vulnerability regardless of whether an error arises from cognitive limitations, ambiguous design or an ordinary software defect.
Respect individual agency. A person may legitimately revise a preference after seeing an alternative. A good system makes that change visible and voluntary. The objective is not to freeze people into their first answers, but to ensure that the account of what they selected and why is neither silently replaced nor inferred from a polished later explanation.
What we still do not know from one striking phenomenon
No single cognitive process explains every failure to detect a substituted outcome. Attention, memory, expectation, similarity, reasoning style, task pressure and social context may contribute in different proportions. The original experiments demonstrate a phenomenon under specified manipulations; they do not isolate one universal neurological pathway.
The frequency of missed swaps is not fixed. It can vary with whether a choice was difficult or easy, how distinctive alternatives were, how long feedback was delayed, what comparison cues were available and how detection was measured. A rate observed in a particular laboratory cannot be moved to school admissions, medical decisions or consumer contracts without relevant evidence.
Explaining an altered outcome is also not a direct measure of everything a person believes. Present reasons may be generated for reasons of coherence or inferred preference, but researchers need carefully constructed contrasts before naming a cause. Avoid terms such as brainwashing or proof that preferences are imaginary; the evidence is narrower and more useful.
The phenomenon challenges a simplistic equation of intention, action and explanation. It encourages triangulation: inspect behaviour at the time, inspect feedback that was delivered, and listen to current verbal interpretation. All three can supply evidence without any one being treated as infallible.
Independent transfer problem with a worked answer
A fictional organisation asks ten staff members to choose one of two training schedules. A clerical error reverses the displayed choice for four staff members, and two of those four flag the mismatch before confirmation. The other two accept the display, but a later review discovers the error. What are the immediate detection proportion among affected staff, the undetected proportion, and the appropriate denominator?
The immediate detection proportion is two divided by four, or 50 percent of affected staff. The immediate undetected proportion is also two divided by four, or 50 percent. It is not valid to say two divided by ten gives the immediate detection rate among affected staff; ten is the total staff count, including six people who never received a swapped choice.
Now suppose one of the two initially unaware staff later requests the alternate schedule after learning about the clerical error. That subsequent request is a transparent change of preference, not evidence of another missed substitution. The organisation should retain the first selection, the erroneous display, the correction and the voluntary later change as separate events.
The repair is operational rather than psychological speculation: review the affected records, notify the four staff, restore or reconfirm their intended options, correct the software or clerical mapping and keep an audit trail. No staff member should be accused of fabrication merely because they did not notice an inaccurate summary. The example tests whether you can keep the denominator and chronology intact.
A short, useful decision-verification habit
Before accepting an important result, say exactly what you originally selected. Compare that choice with the presented summary using stable descriptors such as option name, date, quantity, recipient or role. If anything differs, ask whether it is a recorded revision, a presentation mistake or a real change in the available options.
Where stakes are high, consult the raw record or a trustworthy receipt rather than relying on memory alone. For a joint decision, ask for the approved version and its source. A polished explanation can help identify what people now consider reasonable, but it is not a substitute for confirming what they actually approved.
When a discrepancy appears, investigate without adversarial language. The person may have changed their mind, the interface may have changed the display, a record may have been exported with a wrong label, or there may be an authentic misunderstanding. Each possibility requires a different repair and should be tested against observable evidence.
Finish by preserving the corrected outcome and making any further revision explicit. That keeps autonomy intact while reducing the chance that a mistaken display becomes an enduring organisational fact. The central lesson is simple: independent checking of the choice-to-outcome link is more reliable than treating a confident after-the-fact account as an infallible record.
Frequently asked questions
What is choice blindness in one sentence?
Choice blindness occurs when a person fails to detect that an option presented as the outcome of their own choice differs from the option they actually selected, within the specified opportunity to notice. Sometimes they also provide reasons for the substituted outcome.
Does it mean people always lie about why they choose?
No. The striking feature is that a person can give a sincere-sounding explanation without detecting the substitution. The research does not establish conscious lying and does not show that all introspection is inaccurate.
How is it different from change blindness?
Change blindness concerns an unnoticed visual alteration; choice blindness concerns a mismatch between a documented personal selection and the outcome attributed to that selection. Shared attention limits do not make the two research questions identical.
How is it different from post-event misinformation?
Post-event misinformation investigates how later information affects reports or memory of an earlier event. Choice blindness manipulates the displayed outcome of a known choice. The processes may interact, but their starting evidence and experimental comparisons differ.
Was 69 percent the detection rate in every choice-blindness study?
No. The 2012 moral-attitudes research reported 69 percent of participants failing to detect at least one of two reversals in its particular task. That is not a universal rate of missed changes or a trial-level percentage for the 2005 face-choice study.
Why do reasons sound convincing after a substitution?
A speaker can evaluate the currently displayed alternative and reconstruct an account using what is visible and generally plausible. Some explanations are accurate, others may be influenced by the display, and the causal process cannot be inferred from fluency alone.
How do you calculate an honest detection percentage?
Count unique manipulated events detected within a defined time window and divide by the number of manipulated events that provided detection opportunities. For participant-level measures, divide the number of people who noticed at least one relevant change by the number of eligible people instead.
What should an interface designer do?
Display a specific selection summary, preserve the original record, use clear labels, allow correction and retain an audit trail for important decisions. Verification should help people check meaningful options rather than demand they memorize technical identifiers.
Can I run a hidden swap experiment in school?
Do not covertly alter real learner choices or grades as an informal demonstration. Human-subjects work involving deception requires ethical safeguards and appropriate review. A transparent fictional example can teach the same record-checking distinction without undermining trust.
What is the most useful transfer lesson?
Separate original selection, later feedback and current endorsement. An unnoticed mismatch is not the same as a legitimate change of mind, and a persuasive later explanation is not a replacement for a trustworthy decision record.
Primary research: sources and precise claims
The founding experimental demonstration is Johansson, Hall, Sikström and Olsson (2005), Failure to Detect Mismatches Between Intention and Outcome in a Simple Decision Task, Science 310, 116–119, DOI 10.1126/science.1111709. The university’s publication record identifies the real experimental task and principal finding. Do not apply its rate to unrelated contexts without data.
The original follow-up on moral attitudes is Hall, Johansson and Strandberg (2012), Lifting the Veil of Morality: Choice Blindness and Attitude Reversals on a Self-Transforming Survey, PLOS ONE 7(9), e45457. The paper’s 69 percent finding is explicitly a study-specific participant-level statement about at least one of two changes.
The further survey study is Hall and colleagues (2013), How the Polls Can Be Both Spot On and Dead Wrong: Using Choice Blindness to Shift Political Attitudes and Voter Intentions, PLOS ONE 8(4), e60554. Read it as a controlled research report from the Swedish election setting, not a recipe or claim about today’s voters.
The figures using 24 participants and three altered trials, the 20-versus-50 group calculation and the ten-staff scheduling scenario are independent invented teaching examples. Their arithmetic is intended to teach unit-of-analysis integrity and mismatch checking; none is represented as a new study or a measured real-world effect.
Useful verified specialist reading routes
For the broader treatment of individuals, groups and organisations, read the live How Intelligence Works hub. This page owns the narrower question of whether an attributed choice outcome matches a traceable earlier selection.
If the challenge is failing to notice changes in the scene itself, follow Change Blindness. For altered recall and misleading prompts after witnessing an event, follow Post-Event Misinformation. Those are separate verified specialist owners.
A final self-check: compute an event-level detection rate without adding unmanipulated trials to its denominator; explain why that fraction cannot identify how many people noticed at least one swap; and describe a record design that preserves both the first selection and a legitimate later change of mind.