Contaminated record · genuine or implanted?
Choice-Supportive Bias
After making a choice, people systematically misremember the chosen option as having more positive features than it did and the rejected option as having more negative features, distorting memory at the moment of retrieval to retroactively justify the decision already made.
Exhibit · the account as remembered
Choice-supportive bias is a memory distortion in which the act of choosing rewrites the record: positive attributes get incorrectly attributed to the chosen option and negative attributes to the forgone one, even when those attributes originally belonged to the opposite option. The effect was experimentally named and established by Mather, Shafir, and Johnson (2000) in Psychological Science, using source-monitoring paradigms across four scenarios (job candidates, blind dates, roommates). A 2017 narrative review by Lind et al. found evidence strongest for misattribution on recognition tests, but weaker for recall tests and other forms of misremembering, and a 2019 dissertation found the effect absent when stimuli resembled ordinary decision-making table formats, leaving the phenomenon's robustness genuinely contested. ← reconstructed, not recorded
How the contamination enters
The dominant account, advanced by Mather, Shafir, and Johnson (2000), is a source-monitoring framework: after a choice is made, knowledge of which option was chosen becomes activated during later memory retrieval. When the source of a feature is uncertain, people rely on the heuristic 'I chose this option, therefore positive features probably belong to it and negative features to the other.' This creates systematic misattributions even for features that never existed in the original options (i.e., new lures). The distortion primarily happens at retrieval, not at encoding—demonstrated by Henkel and Mather (2007), who showed that false reminders about which option was chosen suffice to redirect the bias.
Competing account — An older motivational account rooted in Festinger's cognitive dissonance theory (1957) and Brehm's (1956) postdecision attitude spreading posits that the bias is driven by a desire to reduce post-decisional dissonance: people are motivated to believe they made the right choice and therefore selectively recall supporting information. Mather et al. (2000) argued against a purely motivational account because the effect appeared even for hypothetical, low-stakes choices unlikely to generate strong dissonance, and because it was found for misattributions of entirely new features. However, Mather and Johnson (2000) found that affectively reviewing a decision (but not detached analytical review) amplified the bias in younger adults, suggesting motivational and emotional factors can modulate a fundamentally cognitive mechanism. The 2017 Lind et al. review concludes that neither account alone is sufficient.
Seams the examination found
In all four scenarios, participants' memory attributions were systematically choice-supportive: they over-attributed positive features to the chosen option and negative features to the rejected option, for both correctly and incorrectly attributed items. The asymmetry was larger for positive than negative features.
Mara Mather, Eldar Shafir, Marcia K. Johnson · 2000 · N=142 (job), N=75 (dating), N=77 (roommate I), N=172+207=379 (roommate II choice+rejection); all Princeton/UNC undergraduates.
Older adults showed significantly more choice-supportive distortion than younger adults in most conditions. Younger adults matched older adults' level of distortion only in the affective review condition, suggesting emotional processing amplifies the bias.
Mara Mather, Marcia K. Johnson · 2000 · Younger and older adult community members; exact N not confirmed from retrieved sources.
Participants given false reminders about their choice showed choice-supportive distortion in favor of the option they were told they chose—even when this differed from their actual choice. This demonstrates the bias arises primarily during memory retrieval rather than at encoding.
Linda A. Henkel, Mara Mather · 2007 · ~80 participants per key experiment; multiple experiments in the paper.
Choice-supportive memory bias was present for freely chosen options but absent for assigned options. For assigned options, people used a vividness heuristic that led to misattributions favoring the unassigned alternative—the opposite pattern.
Mara Mather, Eldar Shafir, Marcia K. Johnson · 2003 · Princeton undergraduates; multiple experiments.
Investors systematically recalled returns as higher than achieved; larger memory bias correlated with greater overconfidence and trading frequency. Both distortion and selective forgetting of losses were present. Critically, the bias appeared only for actively chosen investments, not for passively observed ones, paralleling the free-choice vs. assigned-option distinction in laboratory work.
Katrin Gödker, Peiran Jiao, Paul Smeets · 2022 · Multiple studies; exact Ns not confirmed from retrieved sources.
Where provenance breaks
You are evaluating a team's decision to adopt a vendor platform last year. Every person who championed the choice vividly recalls its early advantages and struggles to remember the specific weaknesses that competing platforms had; those who weren't involved in the decision recall the tradeoffs more evenhandedly. This asymmetric confidence in the original champions—rooted in memory distortion rather than new evidence—is choice-supportive bias in action.
Authenticity — mixed. The effect as originally operationalized—source misattribution on recognition memory tests using rich, narrative stimuli—has been replicated across multiple studies by different labs (Henkel & Mather 2007; Mather, Shafir & Johnson 2003; Svenson, Salo & Lindholm 2009). However, the 2017 systematic review by Lind, Visentini, Mäntylä & Del Missier concluded that support is 'solid' only for misattribution measured by recognition, and 'significantly weaker' for the other three proposed types (fact distortion, false memory, selective forgetting) and for recall-based tests. A 2019 dissertation by Lind found that when stimuli resembled the attribute-table format common in decision-making research (rather than the narrative descriptions used in the original studies), the effect did not appear despite high statistical power. No formal pre-registered direct replications of Mather et al. (2000) were identified in the retrieved sources. The financial field results (Gödker et al.) provide convergent validity in naturalistic settings.
Records later found altered
Finance / Behavioral Economics · 2022
Investor memory bias and overconfidence in self-directed trading
Gödker, Jiao, and Smeets (working paper 2022, accepted Review of Financial Studies 2025) documented that investors who actively chose to invest in a risky stock recalled their returns as higher than actually achieved, selectively forgot losses, and formed overly optimistic beliefs that led to excess reinvestment. Critically, participants who merely observed identical investment outcomes without choosing did not show the bias, paralleling the laboratory finding that the effect depends on free choice rather than passive observation.
Finance / Behavioral Economics · 2021
Investor memory of past performance is positively biased and predicts overconfidence
Unpublished PNAS study (2021) by Gödker, Jiao, and Smeets with a different study design: investors recalled their most important trades and then looked up actual returns. In both studies, investors were biased to recall returns as higher than achieved; larger memory biases were associated with greater overconfidence and trading frequency. Study 3 showed overconfidence was reduced when participants looked up trade results instead of recalling from memory.
Medicine / Clinical Decision-Making · 2009
Post-decision consolidation of surgical patient data
Svenson, Salo, and Lindholm (2009) in Judgment and Decision Making reported controlled experiments in which participants chose which of two patients to prioritize for surgery. After the decision, participants' recalled quantitative facts about the patients (e.g., survival probabilities) were systematically distorted to support the choice already made—facts initially favoring the non-chosen patient were adjusted toward the chosen patient's profile.
To authenticate the memory
Record before you decide, then compare against the record after
Mather et al. (2000) found that the distortion occurs during source-attribution at retrieval, not at encoding. Gödker, Jiao & Smeets (2021) showed in Study 3 that overconfidence was significantly reduced when investors looked up actual trade records instead of recalling from memory, directly counteracting the memory-based bias.
Easily confused with
- Cognitive Dissonancemechanistically-linkedFestinger (1957) proposed that post-decisional dissonance drives attitude change to support chosen options; choice-supportive bias extends this to memory distortion, but the source-monitoring account challenges pure motivational explanations.
- Confirmation BiassiblingBoth biases favor information consistent with prior commitments, but confirmation bias operates prospectively on incoming information, while choice-supportive bias operates retrospectively on memories of past options.
- Post-Purchase Rationalizationeasily-confusedCommon popular alias for choice-supportive bias, but technically refers to any post-hoc justification of purchases (including conscious rationalization), whereas choice-supportive bias specifically refers to involuntary memory distortion.
- Rosy RetrospectionsiblingAlso involves remembering past experiences more positively than they were, but applies to experiences in general rather than specifically to the memory of choice options relative to their alternatives.
- Escalation of Commitment (Sunk Cost Fallacy)mechanistically-linkedChoice-supportive bias may feed escalation of commitment: distorted positive memories of past choices make decision-makers more likely to continue investing in failing courses of action.
- Hindsight BiassiblingAlso rewrites memory after an outcome is known, but hindsight bias distorts memory of what one predicted would happen, while choice-supportive bias distorts memory of the attributes of the options themselves.
- Source Monitoring ErrorparentChoice-supportive bias is a specific instance of source monitoring errors (Johnson, Hashtroudi & Lindsay, 1993) in which the 'choice' serves as a schematic source cue that drives systematic misattribution of features to options.