Did I Choose It, or Was I Shown It? The Limits of Choice Blindness

Explaining a preference first requires identifying which option was preferred. Everyday conversation often assumes that this connection is straightforward. Yet in a short computerized choice task, a rejected option can be presented afterwards as the chosen one without the participant immediately reporting the mismatch. The research arrangement called choice blindness separates a preference from the explanation subsequently attached to it. Its useful question is not whether people know nothing about themselves, but under what conditions they use information about their own choices.

Fumihiko Taya, Swati Gupta, Ilya Farber and O’Dhaniel A. Mullette-Gillman examined this issue in a Singaporean sample in 2014. They investigated more than detection of substitutions. They also considered how later liking ratings differed following choices with or without explanations, and whether the pattern remained at follow-up approximately two weeks later. Detection, explanation and subsequent evaluation were consequently not collapsed into a single outcome. [1]

The gap between choosing and seeing the result

Thirty-two people aged nineteen to thirty participated in the initial session. They selected the more attractive face from twenty pairs of female faces. Each pair appeared briefly, after which the choice was reported through the keyboard and certainty about the decision was rated. The task did not represent selecting a partner or friend. It concerned a narrower event: a brief attractiveness choice between two alternatives and subsequent responses about that choice.

Ten trials ended after selection and certainty rating. On five trials, the genuinely selected face appeared again while the participant typed an explanation. On the remaining five, the rejected face appeared as if it had been chosen, and an explanation was requested for that substituted outcome. The three conditions matter because later ratings were not compared only between altered and unaltered choices. An accurate outcome accompanied by an explanation could also be compared with a choice requiring no explanation.

Face-pair order was randomized across participants, but the sequence of choice-only, accurate-explanation and substituted-explanation trial types was fixed. Calling the entire task randomly ordered would overstate the procedure. Because the study examined how detecting one mismatch related to detecting later mismatches, time and prior experience within the task were especially relevant. It should not be read as twenty wholly disconnected moments. [1]

Detection depends on when it is reported

The researchers searched written explanations for statements indicating that the displayed face had not been chosen, and also included a verbal report. Such reports during the task counted as concurrent detection. They occurred on 27 of the 160 substituted trials, or 16.9 percent. Twelve of the thirty-two participants reported at least one substitution concurrently. The distinction between a percentage of trials and a percentage of people is simple but essential here.

Afterwards, participants were first asked about general problems. Later they were told about two possible task arrangements and asked which one they thought they had experienced; everyone had actually received substitutions. These questions produced more retrospective reports or classifications into the substitution group. The authors could not be certain how many such answers reflected genuine recollection rather than guesses, so their main analysis focused on concurrent reports. Not reporting a mismatch at the time and never having noticed it are not statements with identical certainty.

One detection did not make later detection perfect. Some people who caught a substitution still failed to report subsequent ones. Prior detections were strongly associated with detecting more, but did not provide complete protection. The authors also found an association between choice certainty and later detection after the first detection. This narrower analysis does not establish that confidence guarantees accuracy for every decision made by every participant.

What a later rating measured

After the choice task, faces were rated individually in random order for attractiveness on a nine-point scale. The display did not identify the earlier pair or indicate whether a face had been selected. The aim was to collect evaluations in a separate task rather than leave them embedded in the original choice display. Face-pair similarity was rated separately as well. Resemblance between alternatives and the difference in how much they were liked were not treated as the same variable.

The researchers examined a relative difference: the rating of the originally selected face minus the rating of its rejected alternative. This gap was larger after explaining the accurate outcome than after choosing without explanation. It was smaller after explaining a substituted outcome. The pattern was consistent with a shift towards the option participants were led to believe they had selected. But a smaller gap is not necessarily a reversed ranking. The initial favourite can remain ahead while the distance between the ratings narrows. [1]

The study tested complete inversions separately. Although their average frequency looked higher in substituted trials, the difference in inversion rates was not statistically significant. Taking the significant rating-gap result and describing it simply as preferences being reversed would erase that distinction. The difference measure also could not by itself determine whether the apparent chosen option became more attractive, the other option became less attractive, or both contributed to the change.

Moving from minutes to weeks

Minutes separated initial choices from the later rating task on the first day. Twenty-five participants subsequently completed the online follow-up. This was not exactly fourteen days after the first session for everyone: intervals ranged from nine to twenty-nine days. The same faces were again rated in random order. The initial difference among trial types was not significantly present at follow-up. The short-term pattern should therefore not be assumed to establish a lasting change in preference in this study.

The authors discuss several explanations, including focused attention during explanation, beliefs about what was selected and other cognitive processes. These proposals need not be mutually exclusive. The experiment does not settle one mechanism as the definitive cause. Its small sample, particular face-rating task and reduced follow-up participation also limit broader conclusions. The failure of the examined individual-difference measures to significantly predict detection does not prove that every personal characteristic is irrelevant.

The evidence is useful because a request for an explanation need not merely record a past preference without affecting what follows. When an incorrect outcome is presented, an explanation can become attached to it; even an accurate outcome accompanied by explanation can be associated with a different later evaluation. No claim that people are always mistaken about themselves is needed. A more concrete lesson is to preserve choice records, confidence, explanations and subsequent liking as separate information. Investigating their relationships is more informative than treating one as an unchanging substitute for another.

Sources

[1] Taya, F., Gupta, S., Farber, I., and Mullette-Gillman, O. A. (2014). Manipulation Detection and Preference Alterations in a Choice Blindness Paradigm. PLOS ONE, 9(9), e108515. Methods, detection measures, preference differences and follow-up findings directly examined. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0108515

The source is published under a Creative Commons Attribution licence. Independently written explanation; source prose and experimental materials have not been reproduced. Source access and factual review: October 14, 2026.