Authority Bias: Why the Same Advice Weighs More From a Title

Deferring to authority is usually a good bet, which is why the bias is so hard to see. A physician knows more about medicine than a stranger on a forum, and taking the physician's word saves checking. The shortcut becomes a bias when the deference attaches to the signs of authority, the title, the uniform, the confident tone, rather than to the knowledge they are supposed to indicate, and when it keeps working after the content has stopped deserving it.

Carl Hovland and Walter Weiss showed the mechanism in 1951. Participants read identical articles on four topics, each attributed either to a source they rated as highly credible or to one they rated as untrustworthy, and gave their opinions before, immediately after, and four weeks later. At first the high-credibility sources moved opinion far more; four weeks later the gap had largely closed, because readers remembered the argument but had lost track of who made it, a pattern the authors called the sleeper effect. The source was working on how much the argument was believed, not on how well it was understood.

Robert Cialdini, in his 1984 book Influence, placed authority among six principles of compliance and emphasized how little it takes to trigger it. Titles, clothing and trappings such as a large car or an expensive office can each do the work on their own. The cue is processed fast and the content slowly, and in most daily decisions the fast route wins.

The bias reaches its most serious form when authority commands rather than advises. Stanley Milgram's 1963 experiment, described below, showed that an experimenter in a gray coat could keep ordinary adults delivering what they believed were painful shocks to a protesting stranger. Charles Hofling and colleagues found in 1966 that 21 of 22 nurses on night duty began to carry out a telephone order from an unknown doctor to give a patient twice the maximum dose of an unfamiliar drug, though 31 of 33 nurses asked in a questionnaire said they would refuse. What people believe they would do under authority and what they do are different things.

Stanley Milgram published "Behavioral study of obedience" in the Journal of Abnormal and Social Psychology in 1963. Forty men from New Haven, aged 20 to 50 and recruited through a newspaper advertisement, took part in what they were told was a study of memory and learning at Yale. Each was assigned the role of teacher; a confederate played the learner and was strapped into a chair in the next room. The teacher read word pairs and, for each wrong answer, was instructed to press the next switch on a shock generator whose 30 switches ran from 15 to 450 volts, labeled from "slight shock" through "danger: severe shock" to a final "XXX."

No shocks were delivered, but the teacher did not know that. From 300 volts the learner pounded on the wall and then stopped answering. When the teacher hesitated, the experimenter, in a gray technician's coat, replied with a fixed series of prods: please continue; the experiment requires that you continue; it is absolutely essential that you continue; you have no other choice, you must go on. The experimenter had no power over the teacher beyond the setting and his own calm insistence.

Every one of the 40 participants continued to 300 volts. Twenty-six of them, 65 percent, continued to the end of the board at 450 volts. Many showed extreme tension, sweating, trembling and nervous laughter, and most of those who obeyed protested while doing so. Beforehand, Milgram had asked 14 Yale psychology seniors how many of 100 participants would go to the end of the board; their estimates ranged from 0 to 3 percent, averaging 1.2, and colleagues consulted informally expected few if any to go far. Jerry Burger's 2009 partial replication, stopped at 150 volts for ethical reasons, reproduced Experiment 5 of Milgram's series, reported in his 1974 book, rather than the 1963 study described here. In Burger's base condition 28 of 40 participants, 70 percent, were prepared to continue past 150 volts; in Milgram's Experiment 5, 33 of 40, 82.5 percent, had done so, a difference that was not statistically significant.

Milgram's study is a study of obedience, the extreme case of authority bias. The demo on this page tests the milder and far more common case, the same words weighed differently because of the name attached, and takes its design and figure from Hovland and Weiss's 1951 experiment.

The signs of authority are everywhere, and most of them are cheap to produce. The rows below are illustrations, not findings from the studies cited on this page.

The counters separate the authority from the argument and check whether the one actually supports the other.

• Strip the name. Read the claim as if it had come from a stranger and ask whether it would still persuade you. The demo above does this by assigning you a version at random; you can do it yourself by covering the byline.
• Check that the expertise is relevant. A title certifies knowledge in a domain. Outside that domain, the title is just a word.
• Ask for the basis. Real authority can say why. A source that answers a request for reasons with a restatement of its credentials has given you the only evidence it has.
• Notice the trappings. Uniforms, offices, confident delivery and other people's deference are the cues Cialdini describes, and they are available to anyone. Treat them as a reason to look harder, not to stop.
• Decide in advance what you will not do on instruction alone. Hofling's nurses had policies against exactly what they did; a rule remembered in the moment is worth more than a value held in general.

Authority bias is about deference to a source. Its neighbors involve other ways that something other than the evidence decides what is believed.

• It is not the halo effect, in which one good quality, such as attractiveness or warmth, colors judgments of unrelated qualities. Authority bias is the special case in which the quality is status or expertise and the judgment is whether to believe or comply.
• It is not the bandwagon effect, where a belief is adopted because many people hold it. The bandwagon runs on numbers; authority bias runs on rank, and a single credentialed voice can outweigh a crowd.
• It is not automation bias, the over-reliance on a machine's output. The two share a mechanism, deference to a source presumed competent, but automation bias concerns software and instruments rather than people.