What Forty-One Participants Buys
Social priming's best defence is that the instrument was too blunt for the claim. Fifteen years on, nobody has built a sharper one
Key takeaways
- For about fifteen years social psychology reported that small incidental cues changed behaviour. A warm cup made a stranger seem warmer. Words about the elderly slowed people's walking. Recalling a misdeed made them want soap.
- The founding samples were tiny. Thirty people in the walking study, fifteen in each group. Elliot and Niesta's romantic-red paper opened with twenty-seven participants, fifteen of them in one condition and twelve in the other, and reported a gap so wide you would expect to notice it without measuring anything.
- Registered Reports bite. Ninety-six per cent of ordinary psychology papers conclude their first hypothesis was supported, against forty-four per cent of Registered Reports.
- The best case for the defence is about design rather than content: several priming paradigms are robust when each person contributes many trials, and vanish into the noise when the same data are analysed one measurement per person, which is how social priming was almost always run.
- Between 2014 and 2021 the classic effects were tested preregistered, at scale, often on protocols the original authors specified. Professor priming, across twenty-three labs and 4,493 people, came back as no difference the tests could detect.
- Psychology's replication rate is bad and it is not peculiar. Preclinical cancer biology replicated at 46 per cent, experimental economics at 61.
You are twenty and it is 2010, and you have just read the best thing you will read all term. A researcher hands people a cup of coffee in a lift, hot for some and iced for others, then asks them to judge a stranger from a short description. The people who held the hot cup describe the stranger as warmer. Nobody notices anything. The word does its work in both directions at once, physical and social, because underneath they may be the same word.
It is a beautiful result. It is in Science. It has forty-one participants.
Split between the hot cup and the iced one, that is about twenty people on each side. One seminar group against another seminar group, and the warmth of a stranger hanging on which room happened to say the kinder thing that afternoon.
Start with the version its defenders would recognise
The lazy telling is that social priming was a silly idea that got what it deserved. That telling makes the collapse harder to understand, because silliness does not occupy a generation of careful people for fifteen years.
The claim was that much of social behaviour runs unsupervised. You do not decide how fast to walk. You do not decide how warm a stranger seems. Those outputs get computed from whatever is currently active in the system, and what is active can be nudged by things with no bearing on the situation. If that is right, conscious deliberation is partly a commentary track over a process that has already finished.
None of that is unreasonable. Most of what a brain does is unsupervised, and the alternative, that social judgment waits politely to be authorised, is the less plausible position.
The methods were not casual either. In the 1996 walking study, John Bargh, Mark Chen and Lara Burrows separated the experimenter from the measurement. The person who met the participant was blind to condition and never saw the scrambled-sentence task. The timing was taken by somebody else, a graduate student posing as the next participant, with a hidden stopwatch, over a walk of 9.75 metres to a carpet strip. That design anticipated the objection that the experimenter's expectations produced the result.
It also had fifteen people in each condition, and those two facts have to be held together rather than in sequence. Careful blinding of a study that small buys you a coin flip you can trust the measurement of. The care was real and it was spent protecting a design that could not deliver a reliable estimate whichever way the coin landed.
Since this recurs in everything that follows, it is worth putting plainly. Judging whether a coin is weighted by tossing it thirty times is a losing proposition. A heavily weighted coin gives itself away in thirty tosses. A slightly weighted one does not, and a perfectly fair coin will now and then hand you a run of heads that looks like proof. So a study with few people in it can only catch differences that are large, and on the occasions when it reports one, nobody can tell from the result whether they are looking at a real difference or at a lucky run. Researchers call this power, and all it means is whether a study was big enough to have found the thing it went looking for. Most of the founding priming studies were not. The honesty-box eyes that lifted milk payments 2.76 times are a case of [small samples](https://thewhyweproject.com/articles/do-watching-eyes-make-people-behave) set against pooled results, in an article on whether watching eyes make people behave. The prison study at Stanford is a second famous case built on [small samples](https://thewhyweproject.com/articles/the-stanford-prison-experiment-reread), and an article sorts what it showed into three claims.
One thing was never in dispute. Read the word dog and you recognise cat around ten to twenty per cent faster. That effect was first reported in 1971, it has survived every check the field has thought to run, and it was confirmed again recently in a nineteen-language registered report with more than 25,000 participants. Lexical semantic priming is secure. It is also a different paradigm, measuring reading times within subjects, and its security does not transfer to claims about behaviour. Saying so is the same discipline this article is about to demand of everyone else.