Someone usually helps

In about nine of ten filmed public conflicts, someone helps

City cameras in Amsterdam, Cape Town and Lancaster filmed 219 public conflicts, from heated arguments to serious assaults. At least one bystander helped in 90.9% of them.

Evidence One study: 219 CCTV clips from three cities, coded by four trained assistants.

Textbooks teach the bystander effect: the more people who watch an emergency, the less likely any one of them is to act. That finding comes from experiments, and it has held up. But it answers a different question from the one a victim asks, which is whether anybody will help.

In 2020 a team led by Richard Philpot at Lancaster University coded footage from municipal cameras. The operators had been told to record every public conflict, “from the mildest animated disagreements to grave physical violence”. The cameras covered inner entertainment and city-centre districts: 63 clips from Amsterdam, 61 from Cape Town and 95 from Lancaster. In 90.9% of the 219 conflicts at least one bystander did something to help, and on average 3.76 people did. The rate was similar in all three places, though they differ widely in how safe their streets feel. The more people present, the more likely someone stepped in: each extra bystander raised the odds of help by a factor of 1.1.

What counted as help was broad: a calming gesture towards an aggressor, a calming touch, holding an aggressor back, consoling a victim, or practical help to someone hurt. The footage had no sound, so spoken help, such as saying “I’m calling the police”, could not be coded. In a check on 85 of the clips, consoling was the only act of help in just one.

Each individual can still be less likely to act in a crowd. Someone, though, usually does.

Philpot, Liebst, Levine, Bernasco and Lindegaard, “Would I be helped? Cross-national CCTV footage shows that intervention is the norm in public conflicts”, American Psychologist 75, 66–75 (2020).

A fair coin lets you down after heads

In a short run of fair flips, the average share of heads that follow a head is below half. That counting trap is why a famous 1985 “no hot hand” result was reopened in 2018.

Evidence The coin counts are exact arithmetic, no sample. The basketball re-analysis uses 26 players from one 1985 experiment and is still debated.

Flip a fair coin three times and look at the flips that come straight after a head. Six of the eight possible runs have at least one such flip; TTT and TTH have none. Work out the share of heads for each of those six runs and average the six shares: 5/12, about 42%. For four flips, 14 of the 16 runs qualify and the average is 17/42, about 40.5%.

The coin has no memory. The shortfall comes from the averaging: every run counts once, however many flips follow a head in it. HHHT has three such flips and scores 2/3, but it counts no more than HTTT, which has one and scores 0. Pool every flip into a single count instead and the share comes out at exactly half. For three flips that is 4 heads out of 8.

Run it yourself. The tool flips sequences of the length you choose, works out the share for each one that has a head to look at, and averages those shares the way a researcher would. It then pools every flip into one count. Beside the simulation it shows the exact value for the left-hand figure, computed by counting every possible sequence.

In 1985 Thomas Gilovich, Robert Vallone and Amos Tversky studied basketball shooters, found no streak shooting beyond chance, and called belief in the hot hand a fallacy. Their controlled experiment had 26 Cornell players take 100 shots each. It compared each player’s hit rate after three hits with the rate after three misses, a selection that carries the same bias as the coin count above. Joshua Miller and Adam Sanjurjo showed in 2018 that the bias works against finding a streak. Corrected, the average gap between the two rates in that experiment moved from 3 to 13 percentage points (p < .01).

That does not settle the hot hand. In 2022 David Ritzwoller and Joseph Romano applied permutation tests to four controlled shooting experiments. They concluded that substantially larger data sets are needed, and in the experiment whose data they could obtain only one shooter departed significantly from randomness. What survives is the narrower point: the 1985 test could not show what it claimed to show.

Miller and Sanjurjo, “Surprised by the hot hand fallacy? A truth in the law of small numbers”, Econometrica 86, 2019–2047 (2018) (free preprint); Ritzwoller and Romano, “Uncertainty in the hot hand fallacy”, Review of Economic Studies 89, 976–1007 (2022); Gilovich, Vallone and Tversky, Cognitive Psychology 17, 295–314 (1985). The 5/12 and 17/42 figures are exact counts over all possible sequences.

You will remember a word you never read

When students studied a list of related words, they recalled a related word that was never on the list almost as often as the words that were: on 55% of lists, against 62% for the words shown.

Evidence Two laboratory experiments: 36 and 30 Rice University undergraduates (1995).

Read these fifteen words, one at a time, then answer a short test. Do not write them down.

In 1995 Henry Roediger and Kathleen McDermott gave Rice University undergraduates lists like this one, each made of associates of a word that was left out. In the second experiment (30 students, lists of 15 words) the missing word was recalled on 55% of the lists, almost as often as the words that were shown (62%). That is a rate per list, averaged over students and lists. It is not the share of students who recalled one particular word. On a later recognition test the students accepted the missing words about as often as words they had seen. In the first experiment (36 students, six lists of 12 words) the missing word was recalled on 40% of the lists and was later recognised with high confidence. The design goes back to James Deese in 1959.

The memory is built from meaning, not from a recording. The word “sleep” fits so well that the mind supplies it.

Roediger and McDermott, “Creating false memories: remembering words not presented in lists”, Journal of Experimental Psychology: Learning, Memory, and Cognition 21, 803–814 (1995); Deese, Journal of Experimental Psychology 58, 17–22 (1959).

A vivid memory is not a more consistent one

Where you were on 11 September feels sharper than last Tuesday. Asked again weeks later, people gave accounts of it that changed just as much.

Evidence One study: 54 Duke University students in three groups of 18. The thinnest evidence on this page.

On 12 September 2001, 54 Duke University students wrote down how they had first heard about the attacks. They also described an ordinary recent event. Weeks later, 1, 6 or 32 depending on the group, the same students described both again, and two raters counted how many details had stayed the same.

The September 11 accounts changed as much as the everyday ones. What differed was belief. Ratings of vividness, of reliving the moment and of confidence in accuracy fell over time for the everyday memory and stayed high for the September 11 one. Confidence did not follow the changes.

The authors’ summary: such memories are special in how accurate they feel, not in how accurate they are. The study had no independent record of what each student had done, so it measured agreement between a person’s own accounts, which the authors treat as necessary for accuracy but not sufficient. Emotional intensity at the start predicted later confidence, not later consistency.

Talarico and Rubin, “Confidence, not consistency, characterizes flashbulb memories”, Psychological Science 14, 455–461 (2003).

People spot lies only a little better than chance

Across 24,483 judges, lies and truths were told apart 54% of the time, where guessing gives 50%.

Evidence Meta-analysis of 206 documents and 24,483 judges (2006).

Most people think they can tell when someone lies to them. Charles Bond and Bella DePaulo pooled 206 documents with 24,483 judges. People judging in real time, with no special training or aids, were right 54% of the time. The authors call that small but not trivial: a mean accuracy d of roughly .40, at about the 60th percentile of the effects social psychologists have meta-analysed.

Lies were caught less often than truths were believed, which means judges lean towards trust. Accuracy was better when judges could hear the speaker than when they could only see them.

Bond and DePaulo, “Accuracy of deception judgments”, Personality and Social Psychology Review 10, 214–234 (2006) (abstract).

Birth order barely shapes your personality

In 20,186 adults from three countries, firstborns and later-borns did not differ on extraversion, agreeableness, conscientiousness, emotional stability or imagination. A 2025 study of far larger online samples found small differences.

Evidence Three national panels, 20,186 people (2015). Two online samples, 710,797 and 74,920 people (2025).

The eldest is responsible, the middle child a peacemaker, the youngest a rebel. Small studies have long reported such patterns, and they are inconsistent with each other.

In 2015 Julia Rohrer, Boris Egloff and Stefan Schmukle used three national panels: the United States (5,240 people), Great Britain (4,489) and Germany (10,457). They compared siblings within the same family and people across families. For extraversion, emotional stability, agreeableness, conscientiousness and imagination they consistently found no birth-order effect, in any sample or design they could test (imagination was not measured in the US panel).

Two small effects did appear. Test scores fall slightly from firstborn to later-born, and self-rated intellect falls by about a tenth of a standard deviation with each step. Intellect and imagination are the two halves of the fifth Big Five trait, openness to experience, so the authors’ conclusion is limited: no lasting effect on broad personality traits outside the intellectual domain.

A 2025 study qualifies that. Michael Ashton and Kibeom Lee analysed the HEXACO personality questionnaire in two online samples of 710,797 and 74,920 adults, mostly from English-speaking countries. People with more siblings scored somewhat higher on Honesty–Humility and Agreeableness; between only children and people from families with six or more children the gaps were d = 0.30 and 0.36. Within a given family size, birth-order differences were considerably smaller, at most about d = 0.10, with firstborns a little lower than middle and youngest children. A tenth of a standard deviation is hard to notice in anyone you meet.

These samples are volunteers who answered online, HEXACO Agreeableness is not the same scale as Big Five agreeableness, and the data cannot show that family structure causes the differences. The fair summary is that birth order matters little, and family size may matter a little more.

Rohrer, Egloff and Schmukle, “Examining the effects of birth order on personality”, PNAS 112, 14224–14229 (2015); Ashton and Lee, “Personality differences between birth order categories and across sibship sizes”, PNAS 122, e2416709121 (2025).

Brainstorm alone, then meet

In 1958, people working separately produced nearly twice as many different ideas as the same number brainstorming together. A 1987 review found the same direction in 18 of 22 experiments.

Evidence A review of 22 experiments (1987). The 1958 original used 96 Yale students.

Alex Osborn, who popularised brainstorming, claimed in 1957 that “the average person can think up twice as many ideas when working with a group than when working alone”.

The test came the next year. Donald Taylor, Paul Berry and Clifford Block had people brainstorm for 12 minutes either alone or in groups of four. They then pooled the ideas of four solo workers, counting duplicates once, to make “nominal groups”. The nominal groups produced nearly twice as many different ideas as the real groups. Of 22 experiments compared in a 1987 review, 18 found nominal groups ahead in the number of ideas and four found no difference, all four with groups of two.

Michael Diehl and Wolfgang Stroebe then asked why. They found the strongest support for blocking: only one person can talk at a time, so ideas have to wait their turn. The authors did not settle exactly how waiting costs ideas. Fear of judgement reduced individual output but did not convincingly explain the gap between groups and individuals, and coasting was doubted as a main cause. The authors note that they never tested all three explanations in one design. A practical reading is to write ideas down alone first and talk afterwards.

Diehl and Stroebe, “Productivity loss in brainstorming groups: toward the solution of a riddle”, Journal of Personality and Social Psychology 53, 497–509 (1987), which reviews the 22 experiments and quotes Osborn (1957) and Taylor, Berry and Block (Administrative Science Quarterly 3, 23–47, 1958).

Talking to a stranger improves the commute

Commuters told to chat to a stranger rated the journey more positive than those told to keep to themselves, and no less productive by their own rating. Other commuters predicted the opposite.

Evidence Field experiments with 97 train commuters and 75 bus commuters, plus laboratory work (2014).

In 2014 Nicholas Epley and Juliana Schroeder asked commuters on Chicago-area trains and buses to do one of three things: talk to a nearby stranger, stay disconnected, or commute as usual. Those told to connect rated the journey more positive than those told to keep to themselves (d = 0.63 on trains, 0.56 on buses). They did not rate it significantly less productive. Against the commute-as-usual group the train difference was not significant; on buses it was. Separate commuters asked to predict the outcome expected the reverse and thought silence would feel better.

Part of the error is that people underestimate how much strangers want to talk. It also persists because people rarely test it. In a laboratory waiting room, those who were talked to enjoyed it as much as those instructed to talk, so the pleasure goes both ways.

The results come from short encounters in the Chicago area, and productivity was a rating, not a count of work done. They do not say that every stranger will welcome a chat.

Epley and Schroeder, “Mistakenly seeking solitude”, Journal of Experimental Psychology: General 143, 1980–1999 (2014).

Hearing it again makes it feel truer, even when you know better

Repeated statements are rated truer than new ones, including false ones that people can correctly answer.

Evidence Two experiments with 40 Duke undergraduates each (2015), a meta-analysis of 51 studies (2010) and a four-experiment follow-up (2020).

Repetition makes a statement easier to process, and ease feels like truth. Psychologists call it the illusory truth effect. For decades the working assumption was that knowledge puts a limit on it: hearing “The Atlantic Ocean is the largest ocean on Earth” over and over would not make you believe it.

Lisa Fazio and colleagues tested that in 2015. They compared people’s ratings of repeated and new statements with what the same people showed they knew in a later check. Repeated false statements gained in perceived truth even when people knew the right answer. The authors call this knowledge neglect: the stored fact goes unused when the claim feels familiar.

The effect is medium-sized across studies: a 2010 meta-analysis of 51 studies found a mean d of about 0.5 for repeated against new statements. In the first experiment, false statements that people were shown to know were false still gained: 2.93 on a six-point truth scale when repeated, against 2.59 when new. A later study found that rating each statement for truth when first seeing it removed the effect for statements that people had the knowledge to judge.

Fazio, Brashier, Payne and Marsh, “Knowledge does not protect against illusory truth”, Journal of Experimental Psychology: General 144, 993–1002 (2015); Brashier, Eliseev and Marsh, “An initial accuracy focus prevents illusory truth”, Cognition 194, 104054 (2020) (author manuscript); Dechêne, Stahl, Hansen and Wänke, “The truth about the truth”, Personality and Social Psychology Review 14, 238–257 (2010).

Stare at a dot and the world fades

Hold your gaze still and a blurred shape away from the point you look at disappears, though it never moves.

Evidence The fading is an old observation you can repeat yourself. The eye-movement study behind the explanation had 8 subjects (2006).

Keep your eyes on the black dot on the left. Do not look at the blue disc. Stay still for ten to thirty seconds and the disc should fade into the background, then come back when you look away.

Ignaz Troxler described the effect in 1804. When the image on the retina stays fixed, the visual system stops responding to it. Normally tiny involuntary eye movements keep the image shifting. In a 2006 experiment in Neuron, eight subjects reported when a peripheral target faded while their eye movements were recorded. Small movements called microsaccades became less frequent before the target faded and more frequent before it reappeared. Whether microsaccades are the whole explanation is still argued, but the fading itself is not in doubt.

Martinez-Conde, Macknik, Troncoso and Dyar, “Microsaccades counteract visual fading during fixation”, Neuron 49, 297–305 (2006); Costela, McCamy, Macknik, Otero-Millan and Martinez-Conde, PeerJ 1, e119 (2013).

Where each claim comes from

Each finding ends with its reference, linked through the DOI. The line marked Evidence says how many people or studies stand behind it, and they range from 8 subjects in a lab to a meta-analysis of 24,483 judges. Every number on this page was checked against the paper’s abstract or full text, or computed from it, the coin counts and the demo’s reference values are exact, and its simulated figures are estimates. The papers appeared in peer-reviewed journals; where a link goes to a free preprint or author manuscript, the page says so. Well-known failed replications, such as power posing and ego depletion, were left out on purpose. A finding that rests on one experiment says so.

What each finding does not show

• Bystanders. The footage comes from inner entertainment and city-centre districts of three cities, with many bars. Help meant any act from a calming gesture to stopping the aggressor, and spoken help was not coded. It shows that someone helps, not that the help is enough or safe.
• Coin and hot hand. The coin counts are exact. The 13-point figure is the authors’ re-analysis of one 1985 shooting experiment with 26 players, not of games, and a 2022 paper says far larger data sets are needed.
• Word list. The 55% and 40% rates are per list, from Rice University students in 1995. The test on this page is a short recognition version and tells you nothing about your memory in general.
• Flashbulb memories. 54 students at one university, measured by consistency between their own accounts. Later studies of other events are not reviewed here.
• Lie detection. The 54% is an average across 206 documents. It does not describe any one person or setting.
• Birth order. The three panels are from Western countries, personality is self-reported, and the 2025 online samples are volunteers. Neither design shows cause.
• Brainstorming. Quantity of ideas is clear. Quality is harder to measure, and in the 1987 review the results for quality depended on how it was measured.
• Strangers. Short encounters in the Chicago area, with productivity rated by the commuters themselves. Long-run benefit is not shown.
• Illusory truth. Medium-sized across 51 studies (d about 0.5), but the 2015 experiments had 40 students each, and the gap between repeated and new false statements was a third of a point on a six-point scale. In a 2020 follow-up, rating each statement for truth at first sight removed the effect for statements people had the knowledge to judge.
• Fading. The role of microsaccades is disputed, and the eye-movement study had 8 subjects. The fading itself is not.