Hofling Hospital Experiment (1966)

Aim

Charles K. Hofling (1966) created a more realistic study of obedience than Milgram’s by running a field study on nurses. The nurses were unaware they were in an experiment.

The main aim was to investigate obedience to authority in a real-world setting, specifically a hospital.

The experiment tested how far nurses would obey a doctor’s command, even when it could harm a patient. It also compared what nurses said they would do with what other nurses actually did.

Method

An unknown “doctor” (actually a confederate in the experiment) called the hospital and ordered the nurses to administer a dangerously high dose of a (fictional) drug to a patient. The dose was twice the maximum daily limit stated on the drug’s label.

The procedure was a field experiment, meaning a staged situation in a real-world setting. It involved 22 real nurses, none of whom knew they were taking part.

Dr Smith, a member of the research team, phoned nurses on night duty at a psychiatric hospital. He introduced himself as a doctor from the psychiatry department. None of the nurses had met him.

He asked each nurse to check the medicine cabinet for a drug called Astroten.

When the nurse checked, the label gave a maximum daily dose of 10 mg.

Dr Smith then told the nurse to give 20 mg to a patient called ‘Mr Jones’. That was twice the maximum. He said he was in a desperate hurry and would sign the authorization form when he arrived, about ten minutes later.

The call ended when the nurse obeyed, resisted, went to get advice, became upset or could not find the drug. It also ended after ten minutes.

The nurses believed the drug was real. It was a placebo, an inert sugar pill invented for the experiment, so no patient could be harmed.

A hidden observer on the ward stopped each trial at a set point. This happened when the nurse had the medication and approached the patient, began to contact another professional, or reached the ten-minute limit. Nobody received any drug.

The order was built so that obeying meant overriding several safeguards at once. Each one alone should have made a careful nurse hesitate.

Safeguards the Order Bypassed

• Unknown Caller: The nurse had never met the doctor and could not confirm he was genuine.
• Telephone Order: Hospital rules did not allow medication to be given on a telephone instruction.
• Unauthorized Drug: The medicine was not on the ward’s stock list.
• Overdose: The 20 mg dose was twice the 10 mg maximum printed on the box.

Together, these made the order easy to refuse on professional grounds. No specialist knowledge was needed, because the label and the written rules supplied the grounds. Refusal needed only a willingness to challenge a physician. The interviews suggest that was the obstacle.

The Questionnaire Comparison Group

The study also included a comparison group. It was not a true control group, which is treated identically to the experimental group except for the variable being tested.

In another hospital, 21 student nurses and 12 graduate nurses were never telephoned. They read the scenario and said what they would do.

The aim was to compare what nurses say they would do with what the telephoned nurses actually did.

Results

All but one of the 22 telephoned nurses obeyed. The table compares what they did with what the questionnaire group said they would do.

Group	Participants	Gave, or would have given, the drug	Refused, or would have refused
Telephoned nurses (behavior)	22	21 (about 95%)	1 (about 5%)
Questionnaire group (prediction)	33	At most 2 (about 6%)	31 (about 94%)

The two columns mirror each other. Four further findings add detail.

• Questioning: Only one telephoned nurse questioned who “Dr Smith” was and why he was on the ward.
• Rules Broken: The nurses were not supposed to take instructions by phone, let alone exceed the allowed dose.
• Interviews: Of the 21 who complied, about half had not noticed the overdose. The rest assumed a doctor’s dose must be safe.
• Questionnaire Group: The 31 who said they would refuse included all 21 student nurses and 10 of the 12 graduate nurses.

Why Did the Nurses Obey?

Because the study varied nothing, the explanations below are interpretations that fit the nurses’ interviews, not tested mechanisms.

Legitimate Authority and Routine Rule-Bending

A physician’s order is the normal trigger for giving medication, and nurses are trained to respond to it. That is legitimate authority.

Bickman (1974) found that people obeyed a uniformed guard more than a civilian, and that this did not depend on being watched. He concluded that legitimacy was the likeliest source of the guard’s power.

The nurses also described telephone orders as common. A violation that is common in practice stops working as a warning. The study may therefore say less about obedience to an extreme order than about how everyday rule-bending lowers the guard against a dangerous one.

Time pressure added to the effect. Dr Smith was in a “desperate hurry”, and asking a colleague meant delaying an urgent request.

Transferred Responsibility and the Agentic State

Dr Smith promised to sign the authorization himself. That placed responsibility for the dose with him.

Milgram (1974) argued that obedient people enter an agentic state. They see themselves as agents carrying out another person’s wishes, so they no longer feel responsible for the consequences.

The interviews fit this account only in part.

About half of the compliant nurses had noticed the overdose and went ahead. They assumed a doctor’s dose must be safe. That is a judgment about expertise, not a reported transfer of responsibility.

Nobody asked the nurses about responsibility, so reading an agentic state into their answers is an interpretation.

Other nurses never noticed the dose. That points to a failure of checking, not of deference. A plausible authority and a routine task can switch off the check that a label invites.

Hofling Compared With Milgram

Feature	Milgram (1963)	Hofling et al. (1966)	Rank and Jacobson (1977)
Setting	University laboratory	Real hospital wards	Real hospital wards
Authority	Experimenter in the room, in a lab coat	Unknown “doctor” by telephone	A real physician’s name the nurses knew, by telephone
Pressure applied	Four standard verbal prods	A single request with a claim of urgency	A telephoned overdose order
Obedience	65% went to 450 V	21 of 22 (about 95%)	2 of 18 (about 11%)

Hofling’s nurses faced no escalating series of prods. The order was a single request with a stated reason for urgency, yet obedience appeared. That suggests the hospital setting did much of the work that the experimenter’s prods did in the laboratory.

The authority was also distant. It was only a voice on a telephone that the nurse could not check. Yet compliance was higher than in Milgram’s laboratory.

The measures differ, so this contrast is a candidate explanation, not an established one.

Conclusion

Hofling et al. demonstrated that people are very unwilling to question supposed ‘authority’, even when they might have good reason to.

When interviewed later, the nurses said that many doctors habitually gave orders by telephone. Doctors became seriously annoyed if they were refused.

Such obedience was against regulations. Even so, the unequal power between doctors and nurses meant life would be very difficult for a nurse who refused.

Hofling’s study showed how the social pressure brought about by the imbalance of power could lead to a nurse actually putting a patient at risk, rather than disobeying orders.

Evidence Check: Does the Hofling Finding Hold Up?

Verdict: Supported with qualifications.

The direction of Hofling’s finding holds. A substantial share of clinicians go along with an unsafe instruction from a senior person, and steeper hierarchy is associated with less challenge.

The 95% figure depends heavily on the conditions, and the questionnaire contrast cannot be read as a controlled experiment.

The 95% Figure Depends on the Conditions

Level of evidence: Experiment; replications conflict.

Of 22 nurses telephoned by an unfamiliar “Dr Smith”, 21 (about 95%) went to give an unfamiliar drug at twice the labeled maximum dose. That broke hospital rules.

Four studies bear on this claim, and none is a clean replication.

• Hofling et al. (1966): One field situation with 22 nurses and no manipulated comparison condition, so it sits at the weak end of the experimental evidence.
• Rank and Jacobson (1977): Telephoned 18 nurses using a real physician’s name, a familiar drug and free consultation with colleagues. Only 2 were prepared to give the overdose.
• Delaloye et al. (2020): Gave 48 simulated pediatric resuscitation teams an incorrect antiarrhythmic order from a senior-physician confederate. 24 followed it.
• Violato et al. (2022): 21 of 59 students did not directly challenge a senior anesthesiologist persisting with a failing intubation. That measures speaking up, not following an order.

Rank and Jacobson changed three variables at once, so theirs is a partial or conceptual replication. Every rerun changes the situation.

A systematic review and a meta-synthesis (Noort et al., 2019; Morrow et al., 2016) test the hierarchy mechanism, not the compliance rate. So they do not raise the level of evidence.

How far can this be trusted? Hofling’s study was small, from one period and place, and predates preregistration, the practice of publishing a study plan before running it.

Rank and Jacobson also had a small sample. Because three features changed together, it cannot say which one mattered.

Delaloye et al. is the most trustworthy evidence on whether authority orders are followed. It is multi-site and behavioral, with professional teams, although it is simulated and its unit is the team.

The effect is real but conditional, not absent. Half of 48 teams followed a senior confederate’s incorrect order.

Yet where conditions came closer to practice, Rank and Jacobson found only 2 of 18.

In the 1960s field study nearly all nurses began to follow an unverified overdose order. The picture then changed. The one telephone-order replication, with a known physician, a familiar drug and colleagues to consult, found 2 of 18.

A modern multi-site simulation found that half of 48 pediatric resuscitation teams followed a senior confederate’s incorrect drug order. The rate therefore depends heavily on the conditions.

What Nurses Say They Would Do Is an Unreliable Guide

Level of evidence: Experiment; partly replicated.

Questionnaire respondents said they would refuse (31 of 33), while the telephoned nurses did not (21 of 22). Professionals’ confidence that they would act as independent safeguards is therefore not good evidence about their behavior.

Three studies bear on this claim.

• Hofling et al. (1966): The only evidence here pairing stated intentions with observed behavior in one profession. The comparison was non-randomized, so it is weak experimental evidence.
• Mazzocco et al. (2025): 414 online adults predicted responses to Milgram’s study. They expected much less obedience than Milgram’s participants showed, and less for themselves than for others.
• Violato et al. (2021): About 70 health-professional students rated four vignettes. In each, more students than a normal spread would predict rated peers as likely to behave worse than themselves.

Violato et al. (2021) measured no behavior at all.

The Hofling comparison has a confounding variable: the group is mixed up with the method. The questionnaire respondents were different people, mostly students, in a different hospital, with nothing at stake. The reported matching cannot be checked.

The Violato et al. sample was small, mostly female and Canadian. Mazzocco et al. used a much larger sample, but it was online and non-clinical. Neither measured behavior.

Mazzocco et al. gives the most trustworthy evidence for the direction of the effect. Telling half the sample the original 65% figure did not significantly change their predictions about themselves. But its predictors were not the people observed. It shares the confound in Hofling’s contrast.

The evidence supports a qualified version of the claim. Two designs six decades apart find that people predict far more resistance than appears in behavior. Both compare one group’s predictions with another group’s behavior.

The size of the gap for clinicians under real pressure has not been measured cleanly.

Stated intentions understate obedience, both in the 1960s nurses and in a large modern online sample predicting Milgram’s study. The size of the gap for practicing clinicians remains unmeasured under a controlled design.

Strengths

The study has three main strengths, each explained below.

• Ecological Validity: Real nurses did their real job without knowing they were observed, so behavior was natural.
• Replicability: A scripted, standardized procedure could be repeated exactly, although the result did not repeat.
• Comparison Group: The questionnaire group turned an anecdote into a comparison of intention and behavior.

High ecological validity

The study has high ecological validity, meaning its findings apply well to real life, because it took place in a real hospital.

The nurses were doing their normal job and did not know they were participants. They could not tailor their behavior to what the researchers wanted, so there were no demand characteristics, the cues that tell participants what is expected.

There is a limit. The nurses said doctors often gave orders by telephone, so the rule being broken was one that was routinely bent. The study shows how a normalized violation is handled. It says less about obedience to openly outrageous orders.

A real-world setting also costs control. Nobody could hold constant the ward’s workload, the staffing or the nurse’s past experience with similar calls.

Replicability

Another strength is the study’s high level of replicability. Replicability is the ability of other researchers to repeat a study. Repeating it, known as replication, shows whether the result holds.

The procedure was standardized. The “doctor” gave the same scripted instructions to every nurse over the phone.

The end of each call was also operationalized, meaning it was defined by a short, fixed list of outcomes. That gave a clear, objective criterion for scoring each nurse’s response.

Other researchers can therefore follow the same procedure and criteria.

A replicable procedure does not guarantee a replicable result, however. The one real replication found a very different figure, as the failed-replication section below shows.

Comparison group

The questionnaire comparison group is a real strength. It turns a striking anecdote into a comparison between what nurses say they would do and what they actually did.

The two samples were reported to be closely matched on age, sex, marital status, length of working week, professional experience and area of origin. Matching helps reduce the influence of individual differences on the results.

It has limits, covered under Weaknesses.

Weaknesses

The study has four main weaknesses, each explained below.

• Comparison Group: The questionnaire group differed in setting, sample and measure, so it was not a true control.
• Sample and Period: Twenty-two nurses in the 1960s United States cannot show what caused the obedience.
• Ethics: Deception, no consent and some distress, although patients were protected.
• Failed Replication: Far less obedience appeared when three conditions changed.

Comparison Group, Not a True Control

The questionnaire group differed from the telephoned nurses in more than the one variable of interest. It was not a true control group.

They worked in another hospital, and most were students rather than qualified nurses. They faced no real caller and no time pressure. The groups were not randomly assigned.

The measures also differed. The telephoned nurses showed behavior, whereas the questionnaire group gave a stated intention. Any difference therefore mixes the situation with the method of measurement.

Two biases probably work in the same direction. People answering questionnaires tend to give flattering answers, known as social desirability. A group not facing the call also feels no pressure to resist.

The gap still shows that professionals’ confidence that they would act as independent safeguards is poor evidence about their behavior. The study cannot show how large the gap is.

The unequal group sizes (22 and 33) are sometimes mistaken for attrition, or dropout. They reflect two different samples, not participants leaving the study.

Sample, Period and Internal Validity

The sample was 22 nurses in the United States in the mid-1960s. Nursing was then overwhelmingly a female profession and medicine overwhelmingly male. The participants’ sex is not reported, so the study cannot separate professional deference from gendered deference to men.

Hospital systems have also changed. Written and electronic orders, pharmacist checks and read-back rules all alter what a telephone order means today.

With no manipulated variable, the study cannot attribute compliance to authority as such. A nurse might have complied for several reasons.

• Authority: The caller was a physician.
• Routine: Telephone orders were normal on the ward.
• No Help: Nobody was available to ask.
• Assumed Reason: The dose might be justified by something the nurse could not see.

The interviews came after the event. Hindsight and self-justification may color them. The samples are small, and the percentages are only descriptive.

A pre-registered design that varied each factor separately would let these explanations be tested.

Ethical issues

The study broke the ethical guideline of deception, because the nurses did not know the “doctor” was a researcher.

Deception was essential. A nurse who knew the call was a test would have behaved differently.

The nurses also gave no informed consent and could not withdraw from a situation they did not know was a study. Some were left distressed. They lacked full protection from harm.

The study put the nurses in a highly stressful conflict between a doctor’s order and hospital rules. That stress could have had negative psychological consequences. Learning they had been willing to give a double dose could also threaten their professional self-image.

Some safeguards were in place.

• Placebo: The drug was harmless, so no patient could be hurt.
• Hidden Observer: An observer stopped every nurse before any drug reached a patient.
• Debriefing: Each nurse was interviewed within about half an hour of the call.

Debriefing does not fully remove the harm that deception and stress can cause.

The balance is a cost-benefit judgment. The knowledge gained, that hospital hierarchies can override a trained nurse’s safeguards, matters for patient safety. The deception and distress would not pass a modern ethics committee without strict conditions.

Failed Replication (Rank and Jacobson, 1977)

Rank and Jacobson (1977) tried to replicate Hofling’s study and found far less obedience. Their result is the strongest evidence that the 95% figure depends on the conditions.

Three things changed at once: a known physician, a familiar drug and access to colleagues. The study therefore cannot say which change explains the drop.

The sample was also small.

Because the conditions changed, this was a conceptual replication, not a direct one. It shows that the finding depends on conditions. That is not the same as low reliability, the consistency of results when a study is repeated under similar conditions.

The fairest reading is that Hofling captured what happens when a nurse cannot judge the drug or verify the caller and must act under pressure. Rank and Jacobson captured what happens when the nurse knows the drug and the physician and can consult colleagues. Both situations arise in real hospitals.

Contemporary Research

The literature reviewed holds no direct modern field replication of the telephone-order design. Since 2015, researchers have studied speaking up through syntheses, surveys and a few simulations.

Hierarchy and Silence Across Healthcare

A systematic review covered 48 studies of speaking up, or “safety voice”, across healthcare, aviation, construction, mountaineering and high-risk sport. It integrated 256 antecedents (contributing factors) (Noort et al., 2019).

A meta-synthesis of 11 qualitative studies from 11 countries and regions covered 504 healthcare workers. It found that hierarchies and power dynamics negatively affect safety voice (Morrow et al., 2016).

Both are syntheses of mostly self-report studies. They describe what staff report more than what staff do. They converge.

Do Teams Follow a Senior’s Wrong Order?

Delaloye et al. (2020) ran a simulated pediatric resuscitation with 48 interprofessional teams. A senior-physician confederate gave an incorrect order. Exactly half the teams followed it.

Following the order was linked to poor challenge within the team and to the confederate’s more hierarchical demeanor. This was a simulation scored at team level, so the link is an association.

Challenging a Senior Doctor: Violato et al. (2022)

The design is among the closest modern behavioral analogues to Hofling’s study, with random allocation to the writing task and two independent coders.

Its limits are clear. The sample was small and came from one institution. The actors were inconsistent, sometimes dismissive instead of authoritative.

The study cannot show what clinicians do on a real ward. It does show that the pattern Hofling reported has not disappeared with better training.

Predictions Stay Optimistic

The gap that Hofling’s questionnaire group revealed has been found again in modern samples asked what they would do.

Health-professional students given short vignettes expected themselves to behave better than the average peer, a better-than-average effect (Violato et al., 2021). Separately, 414 online adults who read an account of Milgram’s shock study predicted much less obedience than Milgram’s participants showed (Mazzocco et al., 2025).

Both are self-report designs on non-representative samples. They show that the optimism is widespread, not how large it is for clinicians under real pressure.

References

Bickman, L. (1974). The social power of a uniform. Journal of Applied Social Psychology, 4(1), 47–61. https://doi.org/10.1111/j.1559-1816.1974.tb02599.x

Delaloye, N. J., Tobler, K., O’Neill, T., Kotsakis, A., Cooper, J., Bank, I., & Gilfoyle, E. (2020). Errors during resuscitation: The impact of perceived authority on delivery of care. Journal of Patient Safety, 16(1), 73–78. https://doi.org/10.1097/PTS.0000000000000359

Hofling, C. K., Brotzman, E., Dalrymple, S., Graves, N., & Pierce, C. M. (1966). An experimental study in nurse-physician relationships. The Journal of Nervous and Mental Disease, 143(2), 171–180. https://doi.org/10.1097/00005053-196608000-00008

Mazzocco, P. J., Reitler, K., Little, L., Korte, J., Ridgill, M., & Stalnaker, X. (2025). Milgram shock-study imaginal replication: How far do you think you would go? Current Psychology, 44(12), 11748–11756. https://doi.org/10.1007/s12144-025-07962-1

Milgram, S. (1963). Behavioral study of obedience. The Journal of Abnormal and Social Psychology, 67(4), 371–378. https://doi.org/10.1037/h0040525

Milgram, S. (1974). Obedience to authority: An experimental view. Harper & Row.

Morrow, K. J., Gustavson, A. M., & Jones, J. (2016). Speaking up behaviours (safety voices) of healthcare workers: A metasynthesis of qualitative research studies. International Journal of Nursing Studies, 64, 42–51. https://doi.org/10.1016/j.ijnurstu.2016.09.014

Noort, M. C., Reader, T. W., & Gillespie, A. (2019). Speaking up to prevent harm: A systematic review of the safety voice literature. Safety Science, 117, 375–387. https://doi.org/10.1016/j.ssci.2019.04.039

Rank, S. G., & Jacobson, C. K. (1977). Hospital nurses’ compliance with medication overdose orders: A failure to replicate. Journal of Health and Social Behavior, 18(2), 188–193. https://doi.org/10.2307/2955382

Violato, E., Witschen, B., Violato, E., & King, S. (2022). A behavioural study of obedience in health professional students. Advances in Health Sciences Education, 27(2), 293–321. https://doi.org/10.1007/s10459-021-10085-4

Violato, E. M., King, S., & Bulut, O. (2021). Conformity, obedience, and the Better than Average Effect in health professional students. Canadian Medical Education Journal, 13(1), 55–64. https://doi.org/10.36834/cmej.71970