A machine wired to nothing 

The centrepiece was a grey instrument panel with thirty switches in a horizontal row, labelled in fifteen volt steps from 15 to 450. Above them ran verbal designations, from Slight Shock through Moderate, Strong, Very Strong, Intense, Extreme Intensity Shock and Danger: Severe Shock, ending in two switches marked only XXX. Throwing a switch lit a pilot lamp, produced an electric buzzing sound, and moved a voltage meter. It was engraved with the name of a fictitious instrument company. It was connected to no source of current whatsoever.

The staging around it was equally deliberate. Two men arrived for each session, one a genuine volunteer and one a confederate, a mild middle-aged accountant playing the part of a fellow participant. They drew slips of paper to assign roles. The draw was rigged so that the volunteer was always the teacher and the confederate always the learner. The learner was then taken to an adjoining room, strapped into a chair with an electrode fixed to his wrist, and told that the straps were to prevent excessive movement. He mentioned a slight heart condition in one widely reported version of the procedure.

The teacher was seated at the generator, given a word-pair memory task to administer, and instructed to move one switch higher for every wrong answer. Crucially, he was given a sample shock of 45 volts from a real battery so that he knew the machine did something. From that point the learner's responses came on a fixed schedule of prerecorded protests, so that every participant met the same sequence of complaints, cries and eventual silence, regardless of what he did.

This is a great deal of theatre for a psychology laboratory, and it is fair to ask whether the deception was justified. The methodological answer is that the study measured a decision, and a decision only exists if the participant believes the alternatives are real. A questionnaire asking people what they would do under such circumstances had already been tried, in effect, when Milgram polled experts beforehand, and the answers were wrong. The ethical answer is much less comfortable, and became the subject of a fight that occupies the last chapter.

There is also a residual doubt that honest accounts include. Milgram's own post-session data indicate that a minority of participants expressed some suspicion that no real shocks were being given, and later researchers working with the Yale archive have argued that the proportion who doubted the setup may have been larger than the published account implies. It is not enough to dissolve the results, and it is enough to make anyone cautious about very precise percentages.

For most participants the machine was real, and at the tenth switch the man next door stopped complaining and started demanding to be let out.