Source Confusion: Why You Remember the Fact but Not Who Said It

A memory is not stored with a tag that says where it came from. The source monitoring framework, set out by Marcia Johnson, Shahin Hashtroudi and Stephen Lindsay in 1993, proposes that the origin of a memory is a judgment made at the time of remembering, built from the qualities of the memory itself. Memories of things you perceived tend to carry more sensory detail and more information about place and time; memories of things you imagined or were told carry more record of the thinking that produced them. Usually the judgment is fast and right. It fails when two possible sources leave memories that feel alike, when the judgment is made quickly or under load, or when the content fits a different source better than the real one.

That is why similarity is the central variable. Lindsay's work found more confusion between two speakers of the same age and sex than between a young woman and an older man, and more when a statement fell on a topic both speakers had raised. The same logic covers the everyday cases: an article and the friend who summarized it leave similar traces; a plan you discussed and a plan you carried out leave similar traces; a detail you saw and a detail you were later told about the same event leave similar traces, which is the mechanism behind the misinformation effect in eyewitness memory. The content is retained; the source is reconstructed, and the reconstruction picks the most plausible candidate rather than the true one.

Source confusion also drives the sense that a claim is true because it is familiar. A statement heard from an unreliable source and later recalled without the source has lost the warning that came with it; what remains is a familiar statement, and familiarity reads as truth. Johnson and colleagues' framework treats this as the same failure: the memory kept its content and dropped the qualities that would have identified where it came from.

Age and attention matter. Older adults show larger source errors than younger ones when recognition of the content is matched, and dividing attention during learning hurts source memory more than item memory. Daniel Schacter listed misattribution among his seven sins of memory in 1999 for this reason: it is a routine failure of a system built to keep content rather than provenance, not a sign of a damaged memory.

The source monitoring framework was published in Psychological Bulletin in 1993 by Johnson, Hashtroudi and Lindsay under the title "Source monitoring." It is a review and a theory rather than a single experiment, and it drew together a decade of work on what Johnson had earlier called reality monitoring: the problem of telling memories of perceived events from memories of imagined ones. The paper argued that one set of processes handles every version of the problem, whether the question is perceived against imagined, one speaker against another, or one occasion against another, and that the decision is made at retrieval from the qualities of the activated memory, weighed against what the rememberer knows about how memories from each source usually feel.

The experiments behind the framework varied the similarity of sources and watched errors rise and fall. In work Lindsay described later, participants heard statements from two speakers and were asked afterward who had said what. Misattributions were more common when both speakers were young women than when one was a young woman and the other an older man, and more common for statements on topics both speakers had addressed than for topics only one had raised; in that study the two kinds of similarity added together. Johnson, Foley and Leach had shown in 1988 that imagining a word in another person's voice made it harder later to tell the imagined words from the ones that person had actually said, which is the same principle applied to imagination.

Lindsay and Johnson applied the framework to eyewitness suggestibility in 1989, arguing that misled witnesses do not lose the original memory so much as misattribute a suggested detail to the witnessed event, and Lindsay, Allen, Chan and Dahl showed in 2004 that details from one event intrude into reports of another more when the two events are similar. The framework has been the dominant account of source memory since, and the brain imaging work that followed, reviewed by Mitchell and Johnson in 2009, has largely confirmed that remembering an item and remembering its source draw on partly separate processes.

The demo on this page follows the two-speaker design in reduced form: two sources, overlapping topics, a short delay, a forced choice of source. It cannot vary speaker similarity, because text has no voice.

The pattern appears wherever information arrives from several places and is used later without its origin. The rows are illustrations, not findings from the cited studies.

Source memory cannot be made perfect, but it can be supported at the moment the information arrives and checked at the moment it is used.

• Attach the source on purpose. Saying or writing "according to" when you take something in gives the memory a distinctive quality to be reconstructed from later.
• Treat a sourceless memory as unverified. If you can recall a claim but not where it came from, that is a reason to check it, not a reason to trust it because it feels familiar.
• Record before you discuss. For anything that may matter later, a written account made before talking to others keeps what you observed separate from what you were told.
• Separate similar sources deliberately. When two people brief you on the same topic, note who said what at the time; afterward the topic alone will not tell you.
• Credit before you are sure. Ideas recalled as your own are often someone else's, and it costs little to ask.

Source confusion is the general failure; several named biases are specific cases or specific causes of it.

• It is not the misinformation effect, which is the change in a witness's memory of an event after misleading information about it. Lindsay and Johnson's account treats the misinformation effect as a source confusion, the suggested detail being attributed to the witnessed scene; source confusion is the broader category.
• It is not cryptomnesia, which is recalling someone else's idea or words as a new idea of your own. That is one kind of source error, between another person and yourself, and it is the case the demo's second example describes.
• It is not false memory in the strict sense, which is a memory of something that did not happen. Source confusion usually concerns something that did happen, remembered correctly in content and wrongly in origin, though a confusion between an imagined event and a real one can produce a false memory.