Famous Psychology Experiments, Checked: What Really Happened in 8 Studies Everyone Quotes

Start hereIn Asch's line studies, three in four people went along with a wrong group at least once. Yet 63.2% of all the answers on those trials went against the group. Which is the real result?

These eight studies all happened. The versions people quote are simpler than what the records show. Below, for each one: the popular version, what the study did, what later work found, and what you can take from it.

01The verdicts

01 Give people a guard's role and they turn cruel on their ownCredited to The Stanford Prison Experiment, as often told ⇄Still argued Two 2019 archive studies found guards were briefed and encouraged; the original researchers reject that reading. Evidence ↓

02 Most ordinary people will obey an authority figure who tells them to hurt someoneCredited to Milgram's studies, as often told ◐Partly Obedience held up but depended heavily on the set-up; why is argued over. Evidence ↓

03 Most people go along with the group, even when the group is plainly wrongCredited to Asch's line studies, as often told ◐Partly Pressure swayed just over a third of answers; most stayed correct. Evidence ↓

04 Nobody helps when others are aroundCredited to The bystander effect, as often told ◐Partly Each person helps less with others there, but someone usually helps. Evidence ↓

05 Workers worked harder just because they were being watchedCredited to The Hawthorne studies, as often told ◐Partly Being studied can change behaviour; the lighting result was not in the data. Evidence ↓

06 Teachers' expectations make children smarterCredited to The Pygmalion study, as often told ◐Partly Expectation effects are usually small; effects on intelligence are unclear. Evidence ↓

07 Pavlov trained his dogs to drool at a bellCredited to Pavlov's dogs, as often told ◐Partly Conditioning is real; his lectures name the metronome most, and accounts disagree about the bell. Evidence ↓

08 Children who wait for a second marshmallow do better in lifeCredited to The marshmallow test, as often told ◐Partly The link is real but smaller, and shrinks by two thirds once family background, early ability and the home environment are counted. Evidence ↓

021. The eight at a glance

All eight studies happened; in each, the popular version leaves out part of the result.

Study	The popular version	What we found
Stanford Prison Experiment, 1971	Guards turned cruel on their own	Still argued
Milgram’s obedience studies, 1963	Most people obey orders to hurt someone	Partly
Asch’s line studies, 1950s	Most people go along with the group	Partly
Bystander studies, from 1968	Nobody helps when others are around	Partly
Hawthorne studies, 1924 to 1932	Workers worked harder because they were watched	Partly
Pygmalion study, 1966	Teachers’ expectations raise children’s IQ	Partly
Pavlov’s dogs	The dogs learned to drool at a bell	Partly
The marshmallow test	Waiting for a treat predicts success in life	Partly

032. Roles and authority

The prison and shock studies show that the set-up shapes behaviour; why people acted so is still argued.

“Give people a guard’s role and they turn cruel on their own”

Still argued The Stanford Prison Experiment showed that ordinary people turn cruel when handed power, with nobody telling them to. Archive studies found the guards were briefed and encouraged; the original researchers reject that reading. The study was stopped early; the argument is about what caused the cruelty.

How sure: Unsettled two 2019 archive studies and the original researchers disagree; the archive evidence leans against “on their own”.

The study ran at Stanford University in 1971, and its official website says the planned two-week study was ended after only six days. Philip Zimbardo wrote there that “our guards became sadistic and our prisoners became depressed and showed signs of extreme stress”.

A 2019 study of the archives, with interviews of 15 participants, found that “the guards received precise instructions regarding the treatment of the prisoners”. It also found that the guards were not told they were subjects, and that data were collected in a biased and incomplete way. The study had drawn on a prison exercise that students in one of Zimbardo’s classes ran three months earlier.

A second 2019 archive study concluded that the experimenters “directly encouraged” the guards to adopt their roles and act tough. Researchers call cues like these demand characteristics: hints that tell participants what the experimenter expects.

Zimbardo and Craig Haney, two of the study’s authors, replied in 2020. They called it “a classic, dramatic demonstration” of what can happen when one group holds nearly total power over another in a dehumanizing place like a prison. They said they value engaged alternative views but reject those that are “ad hominem, misleading, inaccurate, and unscientific”. In a statement on the study’s website, replying to criticism published in 2018, Zimbardo wrote: “We did not give any formal or detailed instructions about how to be an effective guard”. He argued that because some guards remained “good guards” throughout, the “cruel guards chose to act on their own initiative”.

In a 2006 study made with the BBC, men were again split at random into prisoners and guards, and this time the guards “failed to identify with their role”. Zimbardo published a comment challenging that study’s validity.

Our reading: the guards’ behaviour can’t be separated from what they were told. We lean that way because two separate archive studies found instructions and encouragement. For example, when a team starts treating customers or newcomers badly, look first at the instructions, the targets and what the people in charge praise.

“Most ordinary people will obey an authority figure who tells them to hurt someone”

Partly In Milgram’s studies, most ordinary people obeyed orders to give painful shocks. The obedience held up, but how many obeyed depended heavily on the set-up, and why is argued over. Our Influence summary covers the debate in more depth.

How sure: Moderate data from 21 versions show obedience varied widely; why people went on is still argued.

In Stanley Milgram’s 1963 study, 26 of 40 participants gave what they believed was the highest shock. Across the 21 versions of the experiment in which obedience meant going to the maximum shock, 43.6% of 740 people did, so how many obey depends heavily on the set-up. A 2006 partial replication, stopped at 150 volts for ethical reasons, found obedience only slightly lower than Milgram’s.

That 2014 synthesis found eight features that changed how likely people were to go on. They included how directive the experimenter was, pressure from others to disobey, and how close the teacher was to the learner.

A 2015 study of Milgram’s archive argues that participants went on because they identified with what they took to be a worthwhile scientific project. That account is at odds with Milgram’s own explanation. A 2020 re-analysis found that in 18 of 23 versions of the experiment, those who fully believed the shocks were real gave lower shocks on average.

Our reading: obedience rose and fell with the details of the situation. For example, before you follow an order you doubt, notice how the set-up steers you: who gives it, and how far you are from those it affects.

043. Groups

Group pressure is real, but in these studies most answers stayed independent and someone usually helped.

“Most people go along with the group, even when the group is plainly wrong”

Partly Asch’s line studies showed that most people conform to the group. Group pressure swayed just over a third of the answers; most answers stayed correct. Both halves of the result are real, but the second is often left out. Psychologists call going along with others in this way conformity.

How sure: Strong Asch’s own figures show both halves, and a review of 133 later studies found conformity in many countries.

Solomon Asch ran the studies at Swarthmore College in the 1950s. Each real participant sat with actors who, on most trials, all chose the wrong line.

Of the answers on those trials, 63.2% went against the wrong majority and 36.8% went along with it. A quarter of participants never went along, against 5% who always did. Three in four went along at least once, and 95% held out at least once.

Many who went along said they had been certain all along that the group was wrong. A 2015 paper reported that most of the answers on the key trials were independent, correct ones.

A 1996 meta-analysis of 133 Asch-type studies from 17 countries found that conformity in the US had declined since the 1950s. Collectivist countries tended to show more conformity than individualist ones.

Our reading: the pull to agree is real, and so is the habit of holding your ground. For example, if you are the only one who sees a problem in a meeting, write your view down before the group speaks, so your own judgement is on record.

“Nobody helps when others are around”

Partly The bystander effect means that in a crowd, nobody helps. Each person is less likely to help when others are present, but in real public conflicts someone usually does. The bystander effect is real for each person; the “nobody helps” part does not hold up.

How sure: Strong a meta-analysis of over 7,700 people and footage of 219 real conflicts support both halves.

In a 1968 experiment by John Darley and Bibb Latané, people overheard what they thought was someone having a seizure. They believed they were alone, or with one or four others they could not see. Believing others were present made them slower to report it.

The idea is often told through the 1964 killing of Kitty Genovese in New York, with 38 neighbours said to have watched and done nothing. A 2007 study of the archive material found no evidence that there were 38 witnesses, that they saw the murder, or that they did nothing. Our Influence summary covers that story in full.

A meta-analysis confirmed that each individual is less likely to help when others are present, but found the effect weaker in dangerous emergencies. A 2020 study looked at CCTV footage of 219 public conflicts in the UK, the Netherlands and South Africa. It found that “in 9 of 10 public conflicts, at least 1 bystander” did something to help.

Our reading: each person tends to wait for someone else, yet a crowd usually holds a helper. For example, if you notice someone in trouble, don’t wait to see whether others act; they may be waiting too.

054. Being watched, and being expected

Being observed or expected to do well can change behaviour, but by less than the famous stories say.

“Workers worked harder just because they were being watched”

Partly At the Hawthorne plant, output rose whatever the lighting did, because the workers knew they were being studied. Being studied can change behaviour, but the famous lighting result is not what the recovered data show. The idea is now called the Hawthorne effect.

How sure: Moderate the recovered lighting data show no remarkable pattern; a 19-study review finds the effect exists but its size is unclear.

The Hawthorne effect is named after studies at Western Electric’s Hawthorne Works, run from 1924 to 1932. The first were lighting tests from 1924 to 1927 in three departments. The story usually told is that every change in light, even making the room dimmer, raised productivity.

The lighting data were never formally analysed and were thought to have been destroyed. In 2009 two researchers recovered and reanalysed them, and found the remarkable patterns described were “entirely fictional”. They did find “hints of more subtle” signs of a Hawthorne effect in the data.

Later tests in a relay assembly room left researchers unsure why output rose: rest periods, shorter hours, wage incentives, the small group, or the special attention the women received.

A 2014 systematic review of 19 studies concluded that taking part in research does change behaviour, “although little can be securely known” about when, how or how much.

Our reading: measuring something may change it a little. For example, if you start tracking your team’s response times, compare with a team you are not tracking before you credit a new tool.

“Teachers’ expectations make children smarter”

Partly Teachers who expect more from a child raise that child’s IQ. Expectation effects are real but usually small, and whether they change intelligence is unclear. A belief that helps bring itself about is called a self-fulfilling prophecy.

How sure: Moderate reviews covering decades of studies agree the effects are usually small and may fade.

In a 1966 study in 18 classrooms, teachers were told that about one in five of their pupils showed unusual potential; the children had been picked at random. Eight months later those children showed greater IQ gains than the rest, mainly among the younger children. Robert Rosenthal and Lenore Jacobson published the book Pygmalion in the Classroom in 1968.

A 1984 meta-analysis of 18 experiments found that the better teachers knew their pupils when they were given the expectation, the smaller the effect.

A 2005 review of 35 years of research found that self-fulfilling prophecies in classrooms do occur, but are typically small and may fade rather than build up. It said whether they affect intelligence “remains unclear”. Powerful effects may occur among students from stigmatised groups. Teachers’ expectations may predict results mostly because they are accurate.

Our reading: expectations nudge more than they transform, and they matter most when people know each other least. For example, a new manager can wait until they have seen someone’s work before deciding what that person can do.

065. Learning and waiting

Conditioning and the marshmallow link are real; the bell is argued over, and the link is smaller than told.

“Pavlov trained his dogs to drool at a bell”

Partly Pavlov taught his dogs to salivate when a bell rang. The conditioning is real; his lectures name the metronome more than any other sound-maker, and a bell three times. This kind of learning is called classical conditioning.

How sure: Unsettled on the bell — accounts of his work disagree, and his 1927 lectures rarely name one; the conditioning itself is not in doubt.

Pavlov won the 1904 Nobel Prize for his work on the physiology of digestion. In the English edition of his lectures, translated by G. V. Anrep in 1927, a dog starts to salivate 9 seconds after a metronome begins to beat.

In our count of the 23 lectures, metronomes come up 163 times and buzzers 37 times; a bell comes up three times. Musical tones, made with tuning forks, organ pipes and other instruments, come up even more often. In one of the three bell mentions, an electric bell rung after feeding failed to create a conditioned reflex.

In 2014 his biographer Daniel Todes wrote that “Pavlov didn’t use a bell”. He says English speakers think he did partly because of a mistranslation of the Russian word zvonok, which he renders as buzzer. Todes adds that Pavlov’s term was “conditional”, not “conditioned”. Not everyone agrees: a 1997 article argued that Pavlov did use a bell, citing a 1906 report of one of his lectures in Science and a film made in his laboratory in 1925–26. So the sources differ: his biographer Daniel Todes says no bell, the lectures mention one a few times, and a 1997 article says he did.

In the lectures, a learned reflex weakens when its signal is repeated a number of times without reinforcement, which Pavlov called extinction.

Our reading: a learned cue keeps its pull only while it keeps being followed by what it signals. For example, if you want the kettle switching on to become your cue to start studying, follow it with study every time.

“Children who wait for a second marshmallow do better in life”

Partly How long a four-year-old waits for a treat predicts success later in life. The link is real but smaller; most of it goes once family background, early ability and home are counted. Psychologists call this kind of waiting delay of gratification.

How sure: Moderate one larger replication found a smaller link, which shrank further once family, early ability and home were counted.

A 1989 summary of the original research looked at 4-year-olds waiting in certain lab tasks. Those who waited longer grew into adolescents with higher school results who coped better with frustration and stress.

A 2018 study, focusing on children whose mothers had not completed college, found the link with later achievement was ‘only half the size of those reported in the original studies and was reduced by two thirds’ once family background, early ability and the home environment were taken into account. Most of the achievement boost linked to waiting came from managing to wait at least 20 seconds. Links with behaviour at age 15 were much smaller and rarely statistically significant.

Our cognitive biases article puts this result beside 13 other famous findings. Our reading: one wait at age four says little on its own; family background, early ability and the home environment account for much of the link. For example, a child who eats the treat at once has not been set on a path.

076. How to check a famous study yourself

Six questions catch most of what changed in these eight stories.

1. Find the original paper, not the retelling. The Hawthorne lighting story spread for decades before anyone reanalysed the data.
2. Read the numbers, not the headline. In Asch’s studies, most answers went against the group.
3. Ask what the participants were told. The Stanford guards’ instructions changed how the study reads.
4. Look for a replication or a later review. The marshmallow and Pygmalion links both shrank in later work.
5. Check whether the archives were reopened. Archive studies changed how the prison and shock studies are read.
6. Check the words. One translated word may be part of why the bell stuck to Pavlov’s story.

A replication repeats a study to see whether the result holds. A meta-analysis pools many studies of the same question. Our guide to reading a study without a science degree has nine questions for any new finding.

08Quick answers

09Use this

10Quick quiz

How we checked this · Last checked October 2026

• We read the abstracts of the studies and reviews through PubMed, OpenAlex, Crossref and Europe PMC, the NBER summary of the Hawthorne reanalysis, Harvard's Baker Library pages on the Hawthorne studies and the study's official website.
• For Pavlov we read all 23 of his lectures in G. V. Anrep's 1927 translation and counted the sounds named; the count is ours.
• The Asch figures come from a 2015 British Psychological Society report of a published review; we could not open Asch's own 1950s papers.
• Our wording on Milgram, Kitty Genovese, the bystander effect and the marshmallow test matches our Influence summary and cognitive biases article.
• Researchers who disagree about a study are reported as a disagreement, with each side's own words.

How we check every article →

11Sources