The ad in the New Haven paper 

Stanley Milgram was 27, newly appointed at Yale, and looking for ordinary people. He did not want students, partly because the psychology of undergraduates was already overrepresented in the literature and partly because a study about obedience to authority reads differently when the participants are men with jobs and mortgages.

So he advertised. Notices went into the New Haven papers and direct-mail invitations followed, offering payment of four dollars plus fifty cents carfare for one hour of participation in a study of memory and learning at Yale University. No experience was needed. The advertisement listed a range of occupations to signal that all comers were wanted, and it named the topic in a way that was both plausible and untrue, because memory was never what was being measured.

The men who answered were clerks, salesmen, teachers, engineers, labourers, mostly between twenty and fifty, from New Haven and the towns around it. In the first published series they were all male. A later variation used forty women and produced a comparable obedience rate, which is worth knowing before anyone builds a theory about gender out of the headline figure.

It is tempting to skip past recruitment to the machine, but the sampling choices carry weight. This was a self-selected group of people willing to answer an advertisement for a paid scientific study in a particular American city in the early 1960s, at a moment when institutional science had unusual prestige and Yale had unusual prestige locally. They were not a random sample of humanity, and Milgram did not claim they were. The strongest honest claim is narrower and still striking: among a broad cross-section of ordinary employed men, given no threat of any kind, compliance with an escalating instruction was far more common than anyone had predicted.

Prediction is the other reason the recruitment matters. Before running the study, Milgram described the procedure to groups of psychiatrists, students and adults and asked them how far people would go. The estimates clustered low, with almost everyone expecting participants to break off early and only a pathological handful continuing. That gap between expectation and result is the actual finding of the programme. The experiment did not reveal that people are cruel. It revealed that we are bad at predicting our own behaviour inside a structured situation.

Each volunteer arrived at an elegant laboratory in Linsly-Chittenden Hall, met a mild-mannered man he believed was another participant, and was shown a piece of equipment that looked like the product of a serious engineering department.