Two photographs, both in grey scale, both of someone you have never met. You point at the one you find more attractive. The researcher takes it back, says nothing, and by the time it returns to your hand it is the other photograph, the one you just rejected. You look at it. Then you are asked, with real interest, why you chose the picture in front of you. Most people answer the question.
🧠 The Study With a Hidden Hand
The paper is called Failure to Detect Mismatches Between Intention and Outcome in a Simple Decision Task, and it ran in Science in 2005. Petter Johansson, Lars Hall, Sverker Sikström and Andreas Olsson built a task with almost nothing in it: two faces, one preference, one spoken reason. Participants chose between pairs of photographs on attractiveness and then explained their choice out loud. In some trials the experimenter used sleight of hand to hand them the face they had not chosen.
A majority of participants failed to notice. They went on to offer specific, fluent reasons for a preference they had not expressed. One man explained that he preferred the picture because he liked blondes, having pointed at the dark-haired woman moments earlier and been handed the blonde. A reason like that can only be invented, because it justifies a choice that never happened.
The authors gave the effect its name: choice blindness.
🤔 The Reason Arrives Late
The usual picture of a decision puts the preference first and the explanation second, as a report on something already settled. Choice blindness suggests the order can invert. The photograph arrives, the mouth opens, and a coherent reason is assembled on the spot from whatever is available. Calling this lying would be wrong. Nobody in the study had a reason to deceive anyone, and the participants sounded entirely sincere. As Timothy Wilson and Elizabeth Dunn put it, introspection works less like a pipeline into our own machinery and more like a narrative built out of whatever happens to be in consciousness at the time, which may or may not match what is actually driving behaviour. The original 1977 paper by Richard Nisbett and Timothy Wilson made the stronger claim that we have little or no introspective access to higher-order thinking. Wilson later conceded that version went too far. The milder form has held up well enough.
Here is the number worth carrying around. In the post-test interviews, 84% of participants said that if a switch had been made in front of them, they would have caught it. The switch was made in front of them. The hand was visible. They were looking at the rejected photograph while defending the one they thought they had picked.
The follow-ups are where the effect stops being a party trick. Shoppers in a supermarket tasted two kinds of jam, said which they preferred, and were then asked to explain their choice while taking more spoonfuls from the rejected pot. There is a tea version. There is a version where participants choose between objects described on slides and the description is quietly altered before they justify it. Paul Eastwick and Eli Finkel found that what people say they look for in a partner does not predict who they actually choose at a speed-dating event, or who they are still seeing a month later. Henkel and Mather got people to believe, through false reminders, that they had chosen options they had never chosen, and the invented choices then attracted the usual protective loyalty in memory.
Two caveats, because the study has earned them. Grey-scale photographs in a lab are not the same as a face across a table, and given time to reflect people do better. Critics such as Alfred Kaszniak have argued that we still have privileged access to some of our own causal data, more than an outside observer does. The defensible claim is narrower than “introspection is worthless”: a stated reason is not automatically a window into the mechanism that produced the choice.
🔗 Why Self-Reports Are Not Evidence
If you ask people why they bought something, why they picked one brand over another, or why they use a feature the way they do, you are collecting constructions rather than mechanisms. The answer arrives quickly, sounds reasonable, and has roughly the same relationship to the underlying process as that blonde preference did. Behaviour leaves different traces: what people click, how long they stay, when they abandon a form, what happens when you change one variable and nothing else. Surveys and interviews are useful for finding hypotheses. Treating them as explanations is where the trouble starts.
This is the same gap that Ernest LaPiere ran into in 1934, when 128 hotels that had served him and his Chinese travelling companions without complaint later wrote back to say they did not accommodate Chinese guests. Words and behaviour are two separate instruments, and they do not always agree.
There is a version of this that matters for anyone building something that explains a person back to themselves. A reading, a profile, an assessment. When the reader says “yes, that is exactly me,” two things could be happening: the text described something real, or the reader just constructed a real-sounding account of it. Much of what gets remembered from that kind of product is the part that held together and felt useful, which is why the older piece on the Barnum effect sits right next to this one.
🎲 Try It on a Friend
Pick something with two near-identical options. Two brands of the same drink, two songs, two photos. Ask your friend to choose, then quietly hand them the other one and ask why they picked it. Most people will answer without hesitating, and the reason will be specific. What makes the trick interesting is not catching anyone out. It is noticing how good those instant reasons sound, including the ones you produce yourself.
One rule: do not run this on anything that actually matters to the person.
The reason is real. It is not always older than the choice.