Confabulation: Why People Explain Choices They Did Not Make

The word carries two senses that overlap without being the same thing. In clinical use, confabulation names false statements or memories produced without the intent to deceive, most often where a memory impairment meets a failure to check what memory returns; it is described in Korsakoff syndrome, in some dementias, and after frontal lobe injury. In the everyday sense it names the reasons people give for their own choices, constructed after the fact rather than read off the process that produced it. What they have in common is that an account of why is assembled rather than retrieved.

Richard Nisbett and Timothy Wilson set out the general case in 1977: people have little direct access to the processes behind their judgments, and apply plausible theories instead. In two of their studies, shoppers evaluated four nightgowns or four identical pairs of nylon stockings. The rightmost item was heavily overchosen, by almost four to one in the stockings study. No participant mentioned position, and asked directly, virtually all denied it, naming the knit, the sheerness or the weave.

That leaves a gap between the cause of a choice and the report about it, and the 2005 experiment below drove a wedge into it by changing the outcome after the choice was made. The report, it turns out, is generated from whatever is in front of the person now.

How closely the clinical sign and the everyday habit are related is unsettled. Asaf Gilboa and Mieke Verfaellie, introducing a 2010 review, describe confabulation as information that is obviously false or clearly inappropriate to the context, given by someone who is unaware that it is false, and report that after a century of work the definition and the number of types are still disputed. One widely used distinction separates provoked confabulation, an ordinary response to faulty memory, from the rarer spontaneous kind seen where amnesia meets frontal damage. Most of the work they survey locates the failure in the processes that evaluate what memory offers, not in a wish to mislead. It is assessed by a clinician, and not by a page like this.

None of this shows that introspection is worthless. People are reasonably good at reporting what they want, what they feel and what they decided. The evidence concerns the processes behind those contents, which are not observable from the inside.

"Failure to detect mismatches between intention and outcome in a simple decision task" appeared in Science in October 2005. Petter Johansson, Lars Hall, Sverker Sikstrom and Andreas Olsson showed 120 participants pairs of photographs of female faces on cards and asked which face in each pair they found more attractive. The experimenter then handed over the chosen card and asked the participant to say why they had chosen it.

Each participant worked through 15 pairs, and 3 of them were manipulated: using a technique from close-up card magic, the experimenter palmed the chosen picture and handed over the rejected one instead. Participants were assigned 2 seconds to decide, 5 seconds, or as long as they liked. Of the 354 manipulated trials, 46, or 13 percent, were noticed at the time, and counting every form of detection, including the post-task interview, no more than 26 percent were ever exposed. Detection did not differ between the 2-second and 5-second conditions, but it was higher when the time was unrestricted, and it was no higher for dissimilar face pairs than for similar ones.

The verbal reports were the striking part. Reports given after a switch did not differ from reports about genuine choices in specificity, in certainty, or in emotionality. Some fell into a category the authors called specific confabulation: reasons that referred to features unique to the face the participant had ended up with and so could not have been reasons for the original choice, such as the hair color of the person they had not picked.

The paradigm has since left the laboratory. Lars Hall and colleagues ran it in a supermarket in 2010 with jam and tea, swapping the jars and tins after people had tasted or smelled and chosen, and most tasters accepted the substituted sample and argued for it. In 2012 the same group built a paper survey of moral opinions that reversed what participants had just written. Sixty-nine percent of participants missed at least one of two changes, and many argued for the opposite of what they had stated minutes earlier.

The everyday kind shows up wherever a reason is asked for and the real cause is not open to inspection. The rows below are illustrations, not study findings.

You cannot introspect your way out of this one, because introspection about processes is the thing that fails. What works is leaving records that do not depend on the story you tell afterwards.

• Write the reason down before the outcome is known. A prediction or a criterion recorded in advance cannot be quietly rewritten once a choice has been made.
• Fix the criteria before looking at the options. The order in which you see things, and which one you saw last, influence choices more than most people's reports allow for.
• Treat confidence in a reason as no evidence that the reason is right: in the 2005 study, reasons for switched choices were as confident and as detailed as reasons for real ones.
• Ask what would have had to be different for you to choose otherwise. A reason that survives that question is doing real work.
• Keep your own dated copy of what you answered, offered or agreed to.
• Expect it in other people's accounts too, including experts explaining their judgment calls. The explanation is data about how the choice looks now, not about how it was made.

If you are worried about memory, about filling gaps without noticing, or about changes in someone close to you, that is a question for a licensed clinician rather than for a demo.

Confabulation is a sincere account that is not the real one. Its neighbors differ in what is wrong, and in what the person knows.

• It is not lying. A lie requires knowing the truth and stating something else; confabulation is produced and believed.
• It is not false memory, which is a remembered event that did not happen. Confabulation can include false memories, but the everyday kind is about reasons rather than events.
• It is not quite misattribution of memory or source confusion, where the content is real and its origin is wrong. The clearest confabulations supply content that was never there, though the clinical literature also counts real memories displaced in time or dropped into the wrong context.
• It is not choice-supportive bias, the tendency to remember a choice you made as better than it was. That improves a real choice in memory; confabulation accounts for a choice that may not have been yours.