Walking Slowly Past the Elderly

In 1996 John Bargh, Mark Chen and Lara Burrows published an experiment that became one of the most cited in social psychology. Participants unscrambled sentences containing words associated with the elderly (Florida, bingo, wrinkle, grey) without the theme being mentioned. Then, having been told the study was over, they walked down a corridor to the lift while a concealed observer timed them.

The primed group walked more slowly.

It is a wonderful result, and you can see why it travelled: it says that a concept activated below awareness can reach straight into motor behaviour, without any intention in between. It became a fixture of popular books, corporate training, undergraduate lectures. It is probably the single best-known finding of the era.

In 2012 Stéphane Doyen and colleagues tried to repeat it, with a larger sample and infrared timing gates instead of a person with a stopwatch. Nothing. No effect of the prime on walking speed.

Then they ran a second version, and this is the part that matters. They told the experimenters what result to expect. Some were led to believe the primed participants would walk more slowly, others the opposite. The effect appeared, in whichever direction the experimenter had been led to expect, when the experimenter was doing the timing by hand.

That is a devastating design, because it does not merely fail to find the effect. It demonstrates a mechanism that would produce the original finding with no priming whatever. And it points at the experimenter rather than the participants, which is uncomfortable, because the original experimenters were not doing anything they would recognise as cheating. They were starting a stopwatch, by hand, while knowing the condition.

The episode became the centre of what was called the replication crisis, and it is worth separating two things it is often taken to show.

The first is a general lesson about measurement: any procedure where a human who knows the hypothesis makes a judgement call is a channel through which the hypothesis can enter the data. This is why blinding exists in medicine, and social psychology had, for a long list of practical reasons, mostly not bothered. The infrared gates are the point of the Doyen study, quite apart from priming.

The second is a specific question about what survived, and here the popular summary overshoots badly. "Priming collapsed" is not true. Semantic and perceptual priming: that reading "doctor" speeds recognition of "nurse" Is robust, replicated for decades, and not in dispute; it is among the most solid results in cognitive psychology. What failed was a particular and much more ambitious class of claim, sometimes called behavioural or goal priming, in which a subtle conceptual cue produces a complex real-world behaviour. Social priming's failures were concentrated there, alongside ego depletion and a handful of other results that had been built into whole popular literatures.

The reason the episode carried such weight is that the failures were not obscure. They were the field's showpieces, the studies in the textbooks and the trade books, and several came from prominent laboratories. It is one thing to learn that some published results do not hold; it is another to learn that the selection process which made a finding famous (surprising, elegant, endlessly quotable) is precisely the process that favours findings which are too good to be true.

Bargh's response at the time was combative, and the exchange was not psychology's finest hour. But the outcome was structural rather than personal. Preregistration, blinded measurement, larger samples, and the multi-lab replication project became normal. A finding is now expected to arrive with its own replication attached, which is exactly what should have been true in 1996 and was not.