Psychology A path through The Default Library ======================================== How minds work, how we know, and which famous findings held up. Step 1. Biological Psychology: The Wet Machine Behind the Mind -------------------------------------------------------------- Why: Neurons, the brain and the body that thinks. Every thought, mood and night's sleep is made of cells that fire or stay quiet, chemicals that drift across a gap twenty nanometres wide, and blood flow that scanners can see. The course teaches what that does and does not explain. .chips button{min-width:0} .row2{display:flex;gap:8px;flex-wrap:wrap} .row2 button{flex:1 1 90px;min-width:0} .legend{font-family:-apple-system,sans-serif;font-size:.8rem;color:var(--muted)} .brainmap circle{cursor:pointer;stroke:var(--ink);stroke-width:1} Strike a funny bone and your arm jerks before you decide anything. Drink a coffee at four in the afternoon and you lie awake at midnight. A blow to the head changes a person's memory or temper. Someone stops taking a medicine and their mood shifts. In each case something happened in tissue, and something happened in experience, and the two cannot be pulled apart. Biological psychology is the course where psychologists meet that fact. Its job is not to explain the mind by pointing at the brain. It is to learn what the tissue does, in what units, at what speed, and, just as important, which popular stories about "the chemical for love" or "the part of the brain for fear" are too neat to be true. The course in one box Who takes itEvery psychology major, usually in year 1 or 2. Neuroscience, nursing, medicine, biology and pharmacy students take a deeper cousin. The British Psychological Society lists biological psychology among its core areas, and the APA guidelines expect undergraduates to link behaviour to its biological basis. What it is forConnecting what people do and feel to the signalling of nerve cells, hormones, and the body that carries them. Comes afterNo psychology prerequisite; school biology helps. Leads toSensation & Perception (35/21), then Cognitive Psychology (35/22), and Clinical Psychology (35/27), where the chemistry of disorders and drugs comes back. What this piece gives youYou will not be able to read a brain scan. You will see how a neuron decides, why "dopamine is the pleasure chemical" is wrong, what a brain image can claim, and how sleep and learning work. Easier first reads: Why You Feel Before You Think (01/01), Sleep Is a Skill (01/04) and Dopamine Is Not the Pleasure Chemical (01/08). This one goes further. Descriptive only, not medical advice. The neuron: a cell that fires or stays quiet The human brain holds about 86 billion neurons, according to a careful 2009 count that replaced the round "100 billion", along with a roughly equal number of supporting cells. A neuron is a cell that collects inputs and, when enough arrive, sends a brief electrical pulse down its fibre. At rest its inside sits about 70 millivolts below the outside. If inputs push it up to about −55 mV, the threshold, channels open, sodium rushes in, and the voltage swings to roughly +30 mV for about a millisecond before resetting. That pulse is the action potential. Alan Hodgkin and Andrew Huxley worked out its mechanism in 1952 on the giant axon of a squid, and shared a Nobel Prize for it in 1963. Two facts matter more than the details. First, it is all or none: a neuron cannot fire a weak spike. A stronger input does not make a bigger pulse, it makes pulses come more often. The rate is the message. Second, the spike travels along the fibre, and fatty myelin wrapped around it makes the signal jump, so speeds reach roughly 100 metres per second in the fastest fibres, against about a metre per second in thin unwrapped ones. C dV/dt = −(V − Vrest)/R + I, and fire when V reaches thresholdthe leaky integrate-and-fire neuron: input charges it, it leaks back, and a spike resets it Interactive · One neuron deciding Steady input (voltage it would reach) A textbook model, not a real cell: time constant 20 ms, rest −70 mV, threshold −55 mV, 2 ms refractory period. Input is the voltage above rest that the cell would reach if it never fired. Below 15 mV of drive, nothing happens, however long you wait. At 20 mV the cell fires about 34 times a second; at 40, nearly 90. The size of every spike is identical. This is the brain's coding trick: not louder, but faster. The synapse: chemical messages, and the myths about them Neurons do not touch. They are separated by a gap about 20 nanometres wide, the synapse, and the signal crosses it as chemistry. Otto Loewi showed in 1921, with two frog hearts, that a nerve releases a substance rather than just a current. When a spike arrives, the cell releases a neurotransmitter, which binds receptors on the next cell and nudges it toward firing (excitation) or away from it (inhibition). Then the leftover is cleared: pumped back (reuptake) or broken down. In the brain, glutamate is the main excitatory transmitter and GABA the main inhibitory one. Pop psychology turns each transmitter into a mood: dopamine is pleasure, serotonin is happiness, oxytocin is love. The truth is messier. Dopamine neurons fire strongly for unexpected rewards and cues, a "better than expected" signal, which is why they matter for learning and wanting, and not simply for enjoying. Most of the body's serotonin sits in the gut, not the brain. Oxytocin does affect social behaviour, but results from squirting it up the nose have been inconsistent, and many early findings shrank as samples grew. One transmitter plays many roles in different circuits; one behaviour uses many transmitters. Interactive · Release, clearing and a reuptake blocker Normal clearing Reuptake blocked (SSRI-style) Presynaptic firing rate Reuptake activity A toy model in arbitrary units: each spike adds one unit; the gap clears at a rate of 10 + 50 × reuptake per second. With full reuptake, a 10 Hz neuron keeps the gap at about 0.16 units; block 80% of reuptake and the level triples to 0.5. That is, roughly, what drugs like SSRIs do within hours. Yet their effect on mood takes weeks, and the slogan "depression is a chemical imbalance" is a simplification that psychiatrists themselves avoid. A 2022 review argued there is no consistent evidence that low serotonin causes depression; others replied that this was never the standard claim and that the drugs still beat placebo on average (a large 2018 comparison of 21 antidepressants found all outperformed placebo, with modest average effects). What can be said: transmitters matter, the story is not one chemical, and no one should change a medicine because of a diagram. Talk to a prescriber. Hormones are the slow cousin. Glands release them into the blood and they act over minutes to days: cortisol, from a chain called the HPA axis, prepares the body for demands; adrenaline acts in seconds. A nerve impulse reaches your foot in a few hundredths of a second; a hormone takes minutes. Two systems, two timescales. The brain's map: what parts do, and what they do not Where things happen in the brain is one of the oldest questions. In 1848 railway foreman Phineas Gage survived an iron bar through the front of his skull, and accounts say his behaviour changed (the details were exaggerated in retelling). In 1861 Paul Broca examined a man who could understand speech but say almost only one syllable; the damage was in the left frontal lobe. In 1953 surgeons removed parts of Henry Molaison's temporal lobes to treat epilepsy; he could no longer form new lasting memories of events, yet he could still learn a new motor skill. Cases like these show that some functions depend on particular tissue. Do not conclude that each part has one job. Functions are networks: damage to one place can break a skill that many places share. Three myths to drop. We do not use 10% of the brain; imaging shows activity throughout, and the brain, about 2% of body weight, uses around a fifth of the body's resting energy. "Left-brained" and "right-brained" people do not exist; a 2013 analysis of over 1,000 scans found no person-wide dominance, although the two halves do specialise a little (language is usually left). And the neuron count is lopsided: the cerebellum, tucked under the back of the brain and often thought of as just for coordination, holds roughly 80% of the neurons. Interactive · If this part is damaged… A cartoon, not an atlas. Pick a region or tap its circle. Deficits are textbook summaries of famous cases and vary by person. Brain imaging: what a scan can and cannot show A scan of the brain is not a photograph of thinking. MRI shows structure. fMRI measures the blood-oxygen signal (BOLD): active neurons use oxygen, blood flow rises a few seconds later, and the scanner sees the change. So it is indirect, slow (seconds) and blurry (millimetres), whereas EEG catches the brain's electrical rhythms in milliseconds but cannot say precisely where they come from. Neither shows a thought, and neither shows cause. For that you need a lesion, a drug, or brain stimulation. The classic warning is a dead Atlantic salmon. In 2009 Craig Bennett and colleagues put one in a scanner, "showed" it photographs of people, and found voxels (3D pixels) that lit up, because with tens of thousands of voxels, each tested at p < .05, some pass by chance. They used it to argue for corrections for multiple comparisons, and it won an Ig Nobel Prize in 2012. Two related traps: "reverse inference" (the amygdala was active, therefore the person felt fear: it is active in many states) and tiny samples. A 2022 study by Scott Marek and colleagues found that linking brain scans to behaviour in the general population needs thousands of people, not the dozens typical of older studies. Interactive · A scan of nothing p < .05p < .01p < .001 No correctionCorrected for all voxels Voxels in the scan Every voxel is pure noise. The grid shows 2,000 of the voxels; the readout counts them all. Bonferroni correction divides the threshold by the number of tests. With 50,000 voxels at p < .05 you expect 2,500 false "activations". Uncorrected, a scan of nothing finds a brain at work. Correcting the threshold for 50,000 tests keeps the chance of any false alarm near 5%. Modern studies correct, share data and replicate; but an old headline of the "this region lights up when…" kind deserves the same suspicion as any small study (see Research Methods in Psychology, 35/19). Sleep: a pressure that builds and a clock that ticks Sleep is not the brain switching off. Over a night it cycles about every 90 minutes through lighter and deeper stages and REM sleep, discovered in 1953 by Eugene Aserinsky and Nathaniel Kleitman, when the brain is highly active and most vivid dreams occur. A 2015 expert consensus recommended that adults regularly sleep at least seven hours; needs vary between people and across life, and no single "right" number fits everyone. The standard picture, proposed by Alexander Borbély in 1982, is the two-process model. Process S is a sleep pressure that rises the longer you are awake and drops during sleep (adenosine is a leading candidate for part of it, which is why caffeine, an adenosine blocker, delays sleepiness without removing the debt). Process C is the circadian clock, set mainly by a tiny region called the suprachiasmatic nucleus and by light, which makes you more alert at some hours than others. You feel sleepy when S is high and C is low. Jet lag and shift work are C and S out of step. Interactive · Pressure against the clock Sleep each night Bedtime Schematic model after Borbély, with rise time 18.2 h and fall time 4.2 h as in early human fits, scale 0 to 1. The circadian curve is a simple wave peaking at 18:00. Illustrative, not a prediction for any person. With 8 hours' sleep and bed at 23:00, pressure at waking falls to about 0.09 of the scale and ends the day near 0.62. Cut sleep to 5 hours and the morning level only gets down to about 0.22: you begin each day with leftover debt, and it does not clear itself overnight. Plasticity: the brain is rewired by use In 1949 Donald Hebb proposed that when one cell repeatedly helps to fire another, the connection between them strengthens. The slogan "cells that fire together wire together" is a later paraphrase. The effect, called long-term potentiation, was shown in rabbit hippocampus by Tim Bliss and Terje Lømo in 1973. It is the leading model of learning: experience changes connection strengths, not the number of cells. Interactive · A Hebbian cell Input A already drives the cell. Input B is too weak alone. Present them together and watch B's connection grow. Present A + BB aloneA aloneReset Rule: when the cell fires, each active input's weight w grows by 0.2 × (1 − w). The cell fires when the summed active weights reach 0.5. Cartoon of the Hebb rule, not a cell model. After three pairings B alone crosses the threshold. The adult brain stays plastic in far larger ways: in a 2000 study, London taxi drivers who had passed the demanding street test had a larger posterior hippocampus than controls, in proportion to years of driving. But claims that new neurons keep forming in adult humans are disputed. Two 2018 papers, one finding almost no new cells in adult hippocampus and one finding plenty, are still argued over. What is solid is that the brain changes with use throughout life, more in youth, and less than the popular word "rewiring" suggests. What a graduate just sees A neuron's message is a rate of identical spikes, not a loudness. Transmitters have many jobs. "The pleasure chemical" and "the chemical imbalance" are slogans. Nerve signals are fast and local; hormones are slow and body-wide. Functions run on networks, so a lesion shows what a region is needed for, not that it holds the skill alone. A brain scan is an indirect, slow, noisy measure and needs statistical care before it says anything. "We use 10% of the brain" and "left-brained people" are myths. Sleep pressure and the body clock are separate and add up to how awake you feel. Learning changes connection strengths, and use shapes the brain across life. Where it shows up next Sensation & Perception (35/21): how receptors turn light and sound into those spikes, and what the cortex does with them. Cognitive Psychology (35/22): memory and attention, and where the lesion cases above matter again. Clinical Psychology (35/27): mood, anxiety and psychosis, medication and therapy, where the transmitter story and its limits come back. Machine learning (34/32): artificial networks borrow the idea of weights strengthened by experience, loosely. In the world: caffeine and shift work, sleep clinics, brain-imaging headlines, anaesthesia, rehabilitation after stroke, and the question of how far any drug can explain a mood. If you want to go deeper 📘 Textbook. James Kalat, Biological Psychology (Cengage, 14th edition). Written for psychology majors; the common choice. 📘 Textbook. Mark Bear, Barry Connors and Michael Paradiso, Neuroscience: Exploring the Brain (Jones & Bartlett / Wolters Kluwer). More depth for neuroscience-leaning readers. 📘 Gentler. Oliver Sacks, The Man Who Mistook His Wife for a Hat (1985): case histories that make the lesion cases human. 🌐 Free course. MIT OpenCourseWare 9.01 Introduction to Neuroscience (Fall 2007). Also the introduction in 9.00SC Introduction to Psychology. The takeaway The mind runs on cells that signal by firing rate, chemicals that cross a gap, hormones that travel slowly, and connections that change with use. That is a powerful picture and a limited one: slogans about single chemicals or single brain regions, and strong claims from small brain scans, need the same scepticism as any psychological finding. Try this: this week, notice a mood shift (after poor sleep, a late coffee or a skipped meal) and think of the two layers: what the tissue is doing, and what you are feeling. Neither explains the other alone. Sources & Further Reading Sources. Action potential and leaky integrate-and-fire model: standard textbook (Kalat; Bear et al.); Hodgkin & Huxley (1952), Nobel 1963. Neuron count: Azevedo et al., Journal of Comparative Neurology (2009): about 86 billion neurons, about 80% in the cerebellum. Loewi (1921). Reward prediction error: Schultz, Dayan & Montague, Science (1997). Antidepressants: Cipriani et al., The Lancet (2018); Moncrieff et al., Molecular Psychiatry (2022), with published replies. Gage (1848), Broca (1861), Molaison (1953, Scoville and Milner). Left/right brain: Nielsen et al., PLoS ONE (2013). Dead salmon: Bennett et al. (2009); Ig Nobel 2012. Marek et al., Nature (2022). REM: Aserinsky & Kleitman (1953). Sleep duration: Watson et al., consensus statement, Sleep (2015). Two-process model: Borbély (1982); Daan, Beersma & Borbély (1984). Hebb (1949); Bliss & Lømo (1973). Maguire et al., PNAS (2000). Adult neurogenesis: Sorrells et al. and Boldrini et al., both 2018. Simulated numbers are toy models computed live. This is descriptive, not medical advice. If you are struggling, a doctor or local helpline (findahelpline.com) can help. 🌐 MIT OCW 9.01 Introduction to Neuroscience 🌐 MIT OCW 9.00SC Introduction to Psychology 📘 Kalat, Biological Psychology 📘 Sacks, The Man Who Mistook His Wife for a Hat Step 2. Learning & Behavior: Surprise Is the Teacher ---------------------------------------------------- Why: Conditioning, reward and how behaviour is learned. Signals learn to predict, actions are shaped by what follows, and the size of the surprise decides how much changes. From Pavlov's dogs to dopamine and the 66-day habit. Your phone buzzes and, before you decide anything, your hand is already moving. A song from your school years comes on and you feel something that has nothing to do with the room you are in. You pull the lever on a vending machine that ate your coins last week, and press it again, hard, as if that would help. None of these needed a thought. They are the footprints of learning that happens without anyone teaching you: a signal that now predicts something, an action that paid off, a reward that came a little less or a little more than expected. Psychologists once thought of this as a rat-and-pigeon topic. It turned out to be the machinery under advertising, games, addiction, parenting, classroom management, and even the training of artificial agents. A graduate of this course keeps one idea above the rest: learning is driven by surprise. When the world matches your prediction, nothing is learned. When it doesn't, the size of the surprise decides how much you change. The course in one box Who takes itEvery psychology degree, usually in year 1 or 2 (the Psychology route here puts it in year 1). Neuroscience, education and animal-behaviour majors take it too; computer scientists meet its twin as reinforcement learning. What it is forExplaining how experience changes behaviour, with rules that were worked out on animals and then shown to hold, with complications, in people. Comes afterNothing: it has no prerequisite. Biological Psychology (35/20) is a good companion. Leads toDevelopmental Psychology (35/24), and beyond it Clinical Psychology (35/27, where exposure therapy rests on extinction), Behavioral Economics (34/46) and Machine Learning (34/32). What this piece gives youYou will not be able to design a training programme. You will see why intermittent rewards hold behaviour so tightly, what an "error signal" is, and why a habit takes longer than a month. Gentler first reads: How Habits Actually Form (02/05) and Dopamine Is Not the Pleasure Chemical (01/08). The toys here are models and one-person demos, not studies. Classical conditioning: a signal learns what it predicts Ivan Pavlov was studying digestion in dogs (work that won him the 1904 Nobel Prize in Physiology or Medicine) when he noticed that the dogs began to salivate before the food arrived, at the sight of the keeper or the sound of his steps. He turned the nuisance into the experiment. A neutral signal, such as a bell, repeatedly came just before food. The food (the unconditioned stimulus) already caused salivation. After enough pairings, the bell (the conditioned stimulus) caused it too. The modern reading is less "association by closeness" and more "the bell became a forecast". The signal has to be informative: if food arrives just as often without the bell, the bell stays meaningless, however often they co-occur (Robert Rescorla, 1968). Biology also sets limits. In a famous 1966 experiment, John Garcia and Robert Koelling found that rats readily link a taste with later sickness, even hours later, but do not as readily link a flashing light and noise with sickness. It is why one bad oyster can end a love of oysters for years. Humans do the same. A brand jingle that predicted a good time, a sound linked to a bad dentist visit, and the sinking feeling at a certain email notification are all classical conditioning. The evidence for fear learning in humans is not the 1920 "Little Albert" demonstration, which had one baby, no proper controls and an unsettled aftermath: researchers still dispute the child's identity and health. Later laboratory work established the effect properly, with consent, and with ways to undo it. That undoing is extinction: present the bell without the food and the response fades. The old association is not erased; it is overlaid by a new, competing one, which is why fear can return after a break and why exposure-based therapies work by repeating safe encounters. (A hint of Clinical Psychology, 35/27.) Interactive · Which kind is it? Classical: a signal comes to predict an outcome that happens anyway. Operant: an action is followed by a consequence that changes how often it recurs. Tap your answer. Question 1 of 4. Operant conditioning: consequences shape what you do next Edward Thorndike (1898) put cats in puzzle boxes and watched them stumble onto the latch, then do it faster each time. He called it the law of effect: an action followed by a satisfying result becomes more likely. B. F. Skinner (The Behavior of Organisms, 1938) built on it, with a box in which an animal's lever pressing was recorded and rewarded under rules the experimenter set. The vocabulary is a trap, so here it is in one table. "Positive" and "negative" mean added or removed. "Reinforcement" and "punishment" mean the behaviour becomes more or less likely. Behaviour more likelyBehaviour less likely Something addedPositive reinforcement: praise after tidyingPositive punishment: a scolding Something removedNegative reinforcement: the alarm stops when you get upNegative punishment: losing screen time Two things graduates know. Negative reinforcement is not punishment; it is escape that rewards. And the effects of punishment are weaker and less reliable than people expect: it suppresses behaviour while it is applied and teaches what to avoid, not what to do. Teaching a new behaviour with a reward works better than ending an old one with a penalty, and animal trainers such as the ones in Karen Pryor's books build whole careers on that. Schedules of reinforcement: when the reward comes shapes the behaviour Not every response is rewarded in daily life, and Skinner and Charles Ferster (1957, Schedules of Reinforcement) found that the pattern of rewards matters as much as the amount. The four classic rules: a fixed ratio (every fifth press), a variable ratio (on average every fifth, but unpredictable), a fixed interval (the first press after ten seconds) and a variable interval (the first press after a random wait around ten seconds). In animal records the typical shapes are: fixed ratio gives bursts and a pause after each reward; variable ratio gives a fast, steady rate with almost no pause; fixed interval gives a scallop, slow after a reward and speeding up as the time nears; variable interval gives a steady moderate rate. Try the rules yourself. This is a person tapping, not a pigeon, and you know what is happening, so your pattern will differ from an animal's. What the graph does honestly show is how often the rule paid out. Interactive · A lever, four rules Fixed ratio 5 Variable ratio 5 Fixed interval 10 s Variable interval 10 s No more rewards Press the leverReset record Press the lever to begin. Reward = a tick on the line. Variable ratio draws each next requirement uniformly from 1 to 9 (mean 5); variable interval draws each wait uniformly from 2 to 18 s (mean 10). "No more rewards" is extinction. Slot machines and many app notifications work as variable-ratio schedules, which is one reason they hold behaviour so tightly. The honest limits: that is an analogy from animal behaviour, not proof that any one product causes addiction, and the size of the effect in people is debated. Another fact graduates know is the partial reinforcement extinction effect: behaviour that was rewarded only sometimes lasts longer once rewards stop than behaviour rewarded every time, because "no reward yet" was always part of the deal. Learning is prediction error: the Rescorla–Wagner idea Why does a bell stop teaching once it predicts the food? Leon Kamin showed in 1969 that if an animal has already learned that light predicts shock, then a tone added to the light is barely learned, though it is paired with the shock just as often. This is blocking. The tone adds no surprise, so nothing is left to learn. In 1972 Robert Rescorla and Allan Wagner wrote the idea down. In words: each trial, the change in how strongly a cue predicts the outcome is proportional to the error, the difference between what happened and what all the present cues together predicted. ΔV = α · (λ − ΣV)change in a cue's strength = learning rate × (outcome that happened − total predicted by all cues present) A worked number: with a learning rate of 0.3 and an outcome of 1, a fresh cue A gains 0.3 on trial one, then 0.3 × 0.7 = 0.21 on trial two, and so on, leaving it near 0.99 after ten trials. Now add cue B alongside A, which already predicts 0.99. The error is only about 0.01, so B gets only about 0.003. Compare B learned beside a fresh A: they share the prediction and each reach 0.5. Interactive · Rescorla–Wagner, trial by trial Learn A Blocking Control (A+B from scratch) Extinction Learning rate Blocking: 10 trials of A, then 10 of A+B. Control: 20 trials of A+B. Extinction: 15 trials of A with reward, then 15 without. Outcome strength λ = 1 when rewarded, 0 when not. The model is famous because it is simple and because it was wrong in useful ways. It cannot explain every effect (latent inhibition, for example), and later models added attention and timing. But the big idea survived and moved: learning follows the gap between expectation and outcome. Dopamine is a teaching signal, not the pleasure chemical In the 1990s Wolfram Schultz recorded single dopamine neurons in monkeys. Early in training they fired when an unexpected juice reward arrived. After training, they fired at the cue that predicted the juice and no longer at the juice itself. If the promised juice failed to arrive, their firing dipped below baseline at the moment it was due. Schultz, Dayan and Montague (1997) showed that this matches a reward prediction error, the quantity in temporal-difference learning, an extension of Rescorla–Wagner to time. The toy below uses that model. Each of six steps after the cue carries a value; the signal is how far each step's reward-plus-next-value differs from what was expected. Train it, then look at where the signal sits. Interactive · Where the error signal goes (model) Training trials Reward arrives Reward omitted Temporal-difference model: learning rate 0.3, discount 0.95, reward 1 after the sixth step. Bars above the line are positive error, below are negative. This is a model of the signal, not a recording. At zero trials, the whole error sits at the reward. After enough trials it has moved to the cue (about 0.74 in this setting, which is the discount to the power six) and the reward produces almost none. Omit the reward after training and the error at that moment is strongly negative. This is why dopamine is better described as a signal for learning and wanting than as "the pleasure chemical", the point of Dopamine Is Not the Pleasure Chemical (01/08). It is a strong, influential account, and not the only one: researchers also link dopamine to motivation, effort and salience, and the field argues over how much of its activity a single prediction-error idea explains. Habits: behaviour that no longer asks what it is for With enough repetition in a stable setting, an action stops being a choice made for its result and becomes a response triggered by the setting. Rodent experiments (Adams and Dickinson, 1981) show it when a reward is made unpleasant and the animal keeps pressing anyway. How cleanly that split between goal-directed and habitual control carries over to people is debated. The better-known number is how long it takes. In a 2010 study, Phillippa Lally and colleagues had volunteers choose a simple daily behaviour and rate how automatic it felt over twelve weeks. The time to reach a plateau of automaticity averaged 66 days, ranging from 18 to 254, and missing one day did not noticeably hurt. "It takes 21 days" has no such basis; it traces to a self-help book, Maxwell Maltz's Psycho-Cybernetics (1960), as commonly told. Interactive · How long until it feels automatic? Fast (18 days) Average (66) Slow (254) Day Model: automaticity = 1 − e−3t/D, where D is the days to reach 95% of the plateau (set to 18, 66 or 254 to match the range Lally et al. reported). The shape follows their curve; your numbers will differ with the behaviour and the person. The practical lesson is the cue. A habit attaches to a setting, so the same action in a new place is a new start, and the easiest habits are the ones with a clear, repeated trigger. Observational learning: you don't have to be rewarded yourself People and many animals learn by watching. In the Bobo doll studies (Bandura, Ross and Ross, 1961), nursery-age children who saw an adult hit an inflatable doll were more likely to hit it themselves than those who saw a calm adult. Bandura's later work stressed that watching someone get rewarded or punished matters (vicarious reinforcement), and that beliefs about one's own ability, which he called self-efficacy, shape what people try. The honest caveats: a doll is not a person, children may guess what the experimenter wants, and the studies show imitation in a lab, not that screens cause real-world violence, which later research addresses with mixed and debated results. What holds up is the basic claim that modelling shapes behaviour. What a graduate just sees A learned response is a forecast: the cue became useful because it predicted something. Learning needs surprise. If the outcome was fully expected, nothing changes. Rewards teach what to do better than penalties teach what not to do. Unpredictable rewards produce the most persistent behaviour; "sometimes" holds tighter than "always". Extinction is new learning laid over the old, so old fear or craving can come back. Dopamine is closer to "this was better or worse than expected" than to "pleasure". A habit is attached to a cue and a setting, and takes weeks to months, not a fixed 21 days. People learn from watching, which is why models, not lectures, change behaviour. Where it shows up next Developmental Psychology (35/24): attachment, play and parenting all run through reinforcement and imitation, and so do the arguments over how much experience can shape a child. Clinical Psychology (35/27): exposure therapy is extinction; behavioural activation is reinforcement; the link of prediction error to addiction and depression is an active research area. Machine Learning (34/32): reinforcement learning (Sutton and Barto) uses the same temporal-difference error, which is why computer scientists and neuroscientists read each other's papers. Behavioral Economics (34/46) and Consumer Behavior (34/58): loyalty schemes, gamified apps and "streaks" are schedules of reinforcement. In the world: animal training, classroom management, product design, addiction treatment and parenting advice (and the folk wisdom to be wary of it). If you want to go deeper 📘 Textbook. Michael Domjan, The Principles of Learning and Behavior (Cengage). The standard research-minded text, strong on Pavlovian conditioning. 📘 Textbook. Paul Chance, Learning and Behavior (Cengage). A more readable alternative, with many everyday examples. 📘 Gentler. Karen Pryor, Don't Shoot the Dog! (1984), on training with rewards, and Sutton and Barto's Reinforcement Learning: An Introduction (MIT Press, free online) for the same ideas as computer science. 🌐 Free course. MIT OpenCourseWare 9.00SC Introduction to Psychology (Fall 2011), which covers learning and conditioning, plus the free Noba chapters on conditioning. The takeaway Behaviour is shaped by what predicts what, and by how much the world differs from the forecast. Signals learn to predict, actions are strengthened by their consequences, the timing of rewards sets how stubborn a behaviour is, and dopamine neurons carry the surprise that drives it all. Habits are what a repeated cue leaves behind, and they take months, not weeks. Try this: this week, notice one cue-and-response pair of your own (the notification and the reach for the phone, the smell and the craving). Write down the cue, the action and the reward. Then change just the cue and see what follows. Sources & Further Reading Sources. Pavlov (Nobel Prize in Physiology or Medicine 1904 for work on digestion); Rescorla 1968, Journal of Comparative and Physiological Psychology 66 (contingency); Garcia & Koelling 1966, Psychonomic Science 4; Watson & Rayner 1920 and later critiques (the child's identity and health are disputed; see Wikipedia's summary of Beck, Levinson & Irons 2009 and Digdon et al. 2014 for the competing identifications); Thorndike 1898; Skinner 1938; Ferster & Skinner 1957; Kamin 1969; Rescorla & Wagner 1972; Schultz, Dayan & Montague 1997, Science 275; Adams & Dickinson 1981; Lally et al. 2010, European Journal of Social Psychology 40 (mean 66 days to plateau, range 18 to 254); Maltz 1960 as the usual source of "21 days"; Bandura, Ross & Ross 1961. Schedule shapes are the textbook animal findings. Every number an interactive shows was computed in node: R–W at rate 0.3: A after 20 trials 0.999; blocking leaves B at 0.014 after ten compound trials; control A and B 0.50 each; extinction A 0.005 after 15 unrewarded trials. TD model: cue error 0.735 (0.956) after 50 or more trials; omitted reward, error −1.00. Habit model at day 21: 97% (D = 18), 62% (D = 66), 22% (D = 254). 🌐 MIT OCW 9.00SC Introduction to Psychology 🌐 Noba Project (free psychology textbook chapters) 📘 Domjan, The Principles of Learning and Behavior 📘 Pryor, Don't Shoot the Dog! This describes research, not medical advice. If you are struggling, a doctor or local service can help; findahelpline.com lists free helplines by country. Step 3. Probability & Statistics: Reasoning with Uncertainty ------------------------------------------------------------ Why: Statistics: the language of every psychology paper. Nothing you measure is exact, and no part comes out the same twice. Statistics is the discipline of knowing how much to trust a number. A machinist turns a hundred steel shafts to "10 millimetres". Measure them with a good micrometer and no two are the same: 9.981, 10.012, 10.004. Nothing is wrong. That scatter is a fact of every process, every material and every sensor. Now stack five of those parts end to end in a gearbox. Will the stack still fit the housing? A pollster calls a thousand people and says 52% will vote yes, give or take three. A drug trial says "significant". A forecast says 30% chance of rain. Every one of these is the same question: given scatter, how much should I trust this number? A graduate of probability and statistics doesn't see a single number any more. They see the shape of the scatter behind it. This piece builds that sight with eight things you can run: dice, tolerances, polls, a drug-trial trap, a belief that updates, and a gearbox stack. The course in one box Who takes it: almost every engineering and science student, usually in the second year; computer scientists meet it again before machine learning. What it's for: deciding what to believe, and how far to trust a measurement, when outcomes are uncertain. Comes after: Calculus I (34/01), which helps with continuous distributions, though the ideas here need little more than arithmetic. Leads to: Machine Learning (34/32) needs it directly. It also shows up in Materials Science & Manufacturing (34/13: process control and defects), Systems Engineering (34/22: reliability) and Computer Security (34/33: randomness and guessing). Honest line: you won't be able to design a clinical trial after this. You will see why "significant" doesn't mean "important", why a bigger sample beats a cleverer guess, and why most real quantities look like a bell curve. For easier first reads, see the library's Probability: How to Not Fool Yourself (12/03), Statistics for Reading the World (12/05) and Bayes' Theorem (12/17); this piece goes further into the engineering side. Random variables: a shape with a centre and a spread A random variable is a number that comes out of a process you can't predict one run at a time. The key move is to stop asking "what will it be?" and ask "what is the shape of what it could be?" Roll two dice and add them: 7 is six times as likely as 2 or 12, so the shape is a tent. Roll many times and the tent appears. Interactive · Roll dice, grow a shape Dice added together Roll 10 timesRoll 1,000 timesClear Without JavaScript: two dice add to 7 with probability 6/36 and to 2 or 12 with probability 1/36 each. The mean total is 7 and the standard deviation is about 2.42. Two numbers summarise a shape. The mean is its balance point: 7 for two dice. The standard deviation (the "spread") says how far values typically stray from the mean: about 2.4 for two dice, which means totals usually land between roughly 5 and 9. The variance is the standard deviation squared, and it is the one that adds up neatly when you combine independent things. standard deviation = √( average of (value − mean)² ) Now the shape that matters most. Measured quantities (shaft diameters, heights, sensor noise) tend to form the normal curve, the bell. Its great convenience is a fixed rule: about 68.3% of values fall within one standard deviation of the mean, 95.4% within two, and 99.7% within three. Engineers use it every day. Say the shaft's spec is 10.00 mm ± 0.05 mm. Slide the process around and see what share of parts get thrown away. Interactive · How many parts fail the spec? Average diameter Spread (standard deviation) Without JavaScript: a centred process with standard deviation 0.025 mm (the spec limits are exactly ±2 standard deviations) scraps about 4.55% of parts. Cut the spread to 0.01 mm and about 0.00006% are scrapped. Shift the average 0.02 mm off-centre with spread 0.02 mm and about 6.7% are scrapped. Two lessons graduates keep. First, spread, not average, decides quality: a process centred perfectly but wide still scraps parts. Second, being off-centre costs a lot even when the spread is fine. That is why factories track both. The central limit theorem: why the bell keeps appearing Why should so many things be bell-shaped? The answer is one of the most useful theorems in mathematics. Take almost any random process, however lopsided, and average a few dozen independent draws. The averages pile up in a bell, centred on the true mean, with spread shrinking like 1/√n. Early forms are due to de Moivre in 1733 (for coin flips) and Laplace, whose general version dates from about 1810. Pick a lopsided source and drag the sample size. Interactive · Average anything, get a bell Flat (0 to 1) Very lopsided Coin (0 or 1) Draws averaged together: n Without JavaScript: for the lopsided source (mean 1, standard deviation 1) the average of n = 30 draws has standard deviation 1/√30 = 0.183, and the 3,000 averages form a close-to-normal bell around 1. At n = 1 you see the raw source. By n = 30 even the lopsided one is a near-bell. The prediction (the curve) uses only two numbers from the source, its mean and its standard deviation. The shape of the source is forgotten. That is why a measurement made of many small independent errors is normal, and why so much of statistics can pretend things are bell-shaped. standard error of an average = σ / √n In words: the average of n draws wobbles less than one draw, by the square root of n. To halve the wobble you need four times the data, not twice. It is the quietest expensive fact in all of experimental science. Sampling and confidence intervals: how sure is a poll? You can't ask everyone, so you ask n people and treat the share saying yes as an estimate. How far off might it be? The standard error from the last section gives the answer. For a share p the wobble is √(p(1−p)/n), and a 95% confidence interval is the estimate plus and minus about two of those. For n = 1,000 and a share near half, that is ±3.1 percentage points: the familiar "margin of error". The wording is subtle, and the interactive makes it concrete. "95% confidence" is a statement about the method, not about one interval: if you ran the procedure again and again, about 95 intervals in 100 would capture the true value. Here the truth is secretly 52%. Run 100 polls. Interactive · 100 polls, one truth People asked in each poll Run 100 polls Without JavaScript: with 1,000 people per poll the margin of error is about 3.1 points; run 100 polls and about 95 of the intervals cover the true value. With 4,000 people the margin halves to about 1.5 points. Watch two things. Roughly five in a hundred intervals miss the truth even when everything is done correctly. And quadrupling the sample from 1,000 to 4,000 only halves the interval. A poll of a country and a poll of a town need the same sample size for the same precision, as long as the sample is random. The real danger is not the margin of error; it is a sample that isn't random at all, which no formula fixes. P-values and Bayes: two ways to weigh evidence A hypothesis test asks whether data are surprising if nothing is going on. The p-value is the probability, assuming nothing is going on, of seeing a result at least as extreme as yours. Flip a coin 100 times and get 60 heads: for a fair coin the chance of a split at least that lopsided in either direction is about 5.7%. It is not the probability that the coin is fair, and it is not the probability that the result was "due to chance". The American Statistical Association issued a rare formal statement on this in 2016, with six principles. Two of them: p-values do not measure the probability that a hypothesis is true, and a p-value does not measure the size or importance of an effect. The classic trap is testing many things. At the usual cut-off of 0.05, a test of a true "nothing" comes out "significant" one time in 20 by luck. Test twenty unrelated things and the chance that at least one lights up is 1 − 0.9520, about 64%. Try it: each press tests twenty effects that are, by construction, nothing. Interactive · Test twenty things that do nothing Run 20 testsReset the tally Without JavaScript: twenty tests of pure noise at the 0.05 level give at least one false "discovery" about 64% of the time, and one in 20 tests on average. If you report only the one that lit up, you have a headline and no finding. This is part of why many published results fail to repeat: in the Open Science Collaboration's 2015 attempt to replicate 100 psychology studies, 97% of the originals had statistically significant results but only 36% of the replications did. Bayes' rule answers a different question: given the data, how should my belief move? Start with a prior belief, multiply by how well each possibility explains the data, and renormalise. Say you test parts and want the defect rate. Know nothing at first (any rate from 0% to 100% equally likely) and add parts one at a time. Interactive · Learn a defect rate from parts Add a good partAdd 10 good partsAdd a defective partReset Without JavaScript: after 20 good parts and no defects, a flat starting belief gives a most-plausible defect rate near 0%, an average of 4.5%, and a 90% credible range of roughly a quarter of a percent to 13%. Twenty clean parts cannot rule out a defect rate near one in eight. Frequentists and Bayesians disagree about the philosophy: one treats probability as long-run frequency and never assigns it to a hypothesis, the other as degrees of belief that data update. In practice both are used, often on the same project, and on big data sets they usually agree. The Bayesian version has one practical advantage here: after twenty clean parts it honestly says "we still can't rule out one in eight". Correlation, regression and the trap of causation Correlation (called r) measures how tightly two quantities move together along a straight line, from −1 to +1. Regression fits the best line through the cloud (by least squares, as in Linear Algebra, 34/04). Both are summaries, and any summary can hide the story. In 1973 the statistician Francis Anscombe made four data sets with, to two decimals, the same mean, the same spread, the same r (0.82) and the same line (y = 3 + 0.5x). Look at each. Interactive · Four data sets, one set of numbers Set 1Set 2Set 3Set 4 Without JavaScript: all four Anscombe sets have mean x = 9, mean y = 7.50, correlation 0.816 and line y = 3.00 + 0.50x. Set 1 is a noisy line, set 2 a curve, set 3 a line with one outlier, and set 4 a single far point that creates the whole slope. The numbers agree and the stories do not: one noisy line, one curve, one line with a rogue point, one vertical stack held up by a single point. Always plot first. The other trap is the famous one. Ice-cream sales and drownings rise together, because summer heat drives both. A confounder (a hidden third factor) makes two things move together without either causing the other. The remedy graduates know is randomised assignment: if a coin decides who gets the treatment, nothing else can be systematically linked to it. Where you can't randomise, causal claims need much more than a line through a cloud. Engineering: tolerance stacks, reliability and Monte Carlo Five parts, each 10 mm ± 0.1 mm, line up in an assembly. What is the worst the total can be out? Worst-case stacking adds the tolerances: ±0.5 mm. It is safe and expensive, because it assumes every part is at its extreme in the same direction at once. If the parts vary independently, variances add, so the standard deviations add in quadrature: root-sum-square, √5 × 0.1 = ±0.224 mm. For five parts that is under half the worst-case figure. When the formulas get hard, engineers use Monte Carlo: let the computer build the assembly ten thousand times with random parts and count how often it fails. Here each part's tolerance is treated as three standard deviations. Choose how many parts are stacked. Interactive · Monte Carlo tolerance stack Parts in the stack Build 10,000 assemblies again Without JavaScript: five parts at ±0.1 mm (three standard deviations) give a worst-case limit of ±0.5 mm, a root-sum-square limit of ±0.224 mm, and in simulation only about 0.3% of assemblies exceed the root-sum-square limit; none come close to 0.5 mm. The honest caveat is that root-sum-square assumes the parts vary independently and are centred. A tool wearing in the same direction on every part breaks that assumption, and so does one supplier's batch. Engineers disagree about when to trust statistical stacking over worst-case; safety-critical designs often insist on worst-case. Reliability follows the same arithmetic. A system that needs all of its parts is a series system, and probabilities multiply: ten components that each work 99% of the time give 0.9910 ≈ 90.4% for the whole. Add a backup in parallel and failure needs both to fail: two units that each work 90% of the time give 1 − 0.1² = 99%. The two rules drive why aircraft duplicate systems and why complicated machines need very reliable components. What a graduate just sees Every number has a spread behind it; the first question is "how scattered?", not "what is it?". Spread, not the average, decides how many parts are scrapped, and centring matters too. Averages of many independent things are bell-shaped, whatever the source looked like. Halving the uncertainty costs four times the data. A 95% interval is a claim about a method: it misses one time in twenty by design. A p-value is "how surprising is this if nothing is going on", not "how likely is it true". Test twenty things and one will look significant; ask how many were tried. Plot the data; correlation is not causation, and only randomising breaks a confounder. Independent errors add in quadrature; series reliabilities multiply. Where it shows up next Machine Learning (34/32): training is fitting a model to noisy data; ideas of overfitting, error bars and Bayes are this course made bigger. Materials Science & Manufacturing (34/13): process capability and scrap rates are the normal curve against a spec limit. Mechanics of Materials (34/11) and Machine Design (34/19): safety factors exist because loads and strengths are distributions, not numbers. Systems Engineering (34/22): reliability budgets multiply series probabilities and add redundancy to push them back up. Computer Security (34/33) and Algorithms (34/24): good random numbers, randomised algorithms, and the odds of guessing a secret. If you want to go deeper 📘 Douglas Montgomery and George Runger, Applied Statistics and Probability for Engineers (Wiley; 7th edition). The standard engineering course text, with plenty of tolerance, quality and reliability problems. 📘 Joseph Blitzstein and Jessica Hwang, Introduction to Probability (CRC Press; 2nd edition, 2019). Best for deep intuition about probability itself, from Harvard's Stat 110. 📖 David Spiegelhalter, The Art of Statistics (2019): the most readable route for learning alone, built around real cases and honest about uncertainty. 🌐 MIT OpenCourseWare 18.05, Introduction to Probability and Statistics (Spring 2022): free notes and problem sets, with Bayesian and classical statistics side by side. The takeaway Nothing you measure is exact, so every number is a shape. Summarise a shape by its centre and its spread, trust averages more as the square root of the sample, read a confidence interval as a promise about a method, treat a p-value as a measure of surprise, and plot before you believe a correlation. Try this: next time you see a poll or a "significant" headline, find the sample size and ask how many things were tested. Then, in the dice widget above, add four dice and watch the tent turn into a bell. Sources & Further Reading Sources. The normal-curve rule (68.3%, 95.4%, 99.7%), the central limit theorem, confidence intervals, the binomial test, the Beta posterior, root-sum-square stacking and series/parallel reliability are standard textbook results (Montgomery and Runger; MIT 18.05); every number shown was recomputed in code. Wasserstein and Lazar, "The ASA's Statement on p-Values: Context, Process, and Purpose", The American Statistician (2016). Open Science Collaboration, "Estimating the reproducibility of psychological science", Science (2015): 97% of originals significant against 36% of replications. Anscombe, "Graphs in Statistical Analysis", The American Statistician (1973). Early forms of the central limit theorem: de Moivre (1733) and Laplace (c. 1810), as surveyed in recent histories such as Ulyanov, "From classical to modern central limit theorems" (arXiv 2405.19828). Go deeper. The books and course above; Wasserstein and Lazar's statement is free and short (doi.org/10.1080/00031305.2016.1154108). Step 4. Research Methods in Psychology: How to Not Fool Yourself ---------------------------------------------------------------- Why: Experiments, measurement, and the replication crisis. Psychology asks questions that cannot be put on a scale, answers them with small noisy samples, and then discovered that many famous answers did not repeat. The methods course is how the field learned to check itself. .chips button{min-width:0} .row2{display:flex;gap:8px;flex-wrap:wrap} .row2 button{flex:1 1 90px;min-width:0} .item{border-top:1px solid var(--line);padding:10px 0;font-family:-apple-system,sans-serif;font-size:.93rem} .item b.nm{display:block;margin-bottom:6px} .item .ans{margin-top:6px;color:var(--ink);background:var(--hi);border-radius:8px;padding:8px 10px;display:none} .item.done .ans{display:block} .legend{font-family:-apple-system,sans-serif;font-size:.8rem;color:var(--muted)} A headline says that people who stand like a superhero for two minutes feel braver and take more risks. It was published in a good journal in 2010, taught in business schools, and watched by tens of millions in a talk. Then bigger studies tried it and the hormone and risk results did not appear. The problem was not a villain. It was a method: small groups, many ways to analyse the numbers, and a system that printed only the exciting result. Every psychology graduate has this story somewhere in their head, and it changes how they read every claim about the mind. They do not ask "is it true?" first. They ask: how was it measured, who was in it, was anyone assigned at random, how big was the effect, and would it happen again? That habit is what a methods course sells. The course in one box Who takes itEvery psychology major, usually in year 1 or 2, alongside statistics. The APA's undergraduate guidelines and the British Psychological Society's accreditation both make research methods a core strand, and neuroscience, nursing, education and sociology programmes teach a version of it. What it is forTurning a question about minds into a measurement, and telling a result worth believing from one that merely happened. Comes afterProbability and Statistics: Reasoning With Uncertainty (34/05). Leads toSocial Psychology (35/25) and Personality & Individual Differences (35/26), where most of the famous disputed findings live. Its cousins are Research Methods in Political Science (35/18), Econometrics (34/44) and Marketing Research and Analytics (34/59). What this piece gives youYou will not be able to design a publishable study. You will be able to see why a small, flexible study can "find" nothing at all, and what the field now does about it. Easier first reads: The Biases That Fool Everyone (17/02) and How to Spot Bullshit (17/03). The statistics here are the ones in 34/05; this piece goes further into what they do in a lab. Measuring what you cannot see: operationalising, reliability and validity You cannot put "stress", "intelligence" or "attachment" on a scale. So the first move of any study is to operationalise: to pick something you can count that stands for the idea. Stress becomes cortisol in saliva, or a score on a questionnaire. Intelligence becomes the score on a set of puzzles. Every such choice is a bet that the number captures the thing, and the bet can be wrong. Two words judge the bet. Reliability is whether you get the same number again: the same person, the same scale, a week later. Validity is whether the number measures what you meant. A bathroom scale that always reads 3 kg too heavy is perfectly reliable and not valid for weight. Reliability is necessary but not enough, and without it nothing else works, because noise in a measure shrinks every relationship you try to find with it. observed correlation ≈ true correlation × √(reliability of X × reliability of Y)called attenuation: noisy measures hide real links Interactive · Noise hides real links Reliability of the measure True link between the two traits Illustrative simulation: 400 invented people, each measured twice with noise; the plot shows 60. Both traits are measured with the reliability you chose. With reliability 0.7 on both measures, a real link of 0.6 shows up as about 0.42. A study of a shaky questionnaire can fail to find a true effect simply because its ruler is made of rubber. Experiments and correlations: only random assignment breaks the confounder People who exercise report better wellbeing. Does exercise cause it? Maybe. Or people with more free time, better health and less money stress both exercise more and feel better: a confounder. A correlational study records what people already do, and cannot tell these stories apart however large it is. An experiment does one thing differently: the researcher decides who gets the treatment, by chance. A coin flip cannot be related to free time, health or anything else, so on average the groups differ in nothing except the treatment. That is why randomised experiments sit at the top of the evidence ladder, and why a study that merely compares people who chose different things should say "associated with", not "causes". Interactive · A link with no cause Observe what people do Assign by coin flip True effect of exercise on wellbeing How strongly free time drives both Invented data, 400 people, standardised scores. Observing: exercise = confounder × c + chance. Coin flip: exercise = chance alone. The plot shows 80 people. With a true effect of zero and a confounder of 0.6, the observed link is about 0.35, and a naive reading says exercise helps. Flip the switch to coin-flip assignment and it falls to roughly zero, which is the truth. Experiments are not always possible (you cannot assign people to grief or to poverty), so psychologists also use careful natural comparisons, but they say so when they have not randomised. Effect size and power: how big, and could this study even see it? "Statistically significant" says only that a result would be unlikely if nothing were going on. It says nothing about size. A huge study can make a trivial difference significant; a small one can miss a large one. So papers now report an effect size. The common one for two groups is Cohen's d: the gap between the group averages, measured in standard deviations. d = (mean of group B − mean of group A) ÷ standard deviationd = 0.5 means the average person in B sits half a standard deviation above the average in A Interactive · What an effect size looks like d = 0.2d = 0.5d = 0.8 Effect size d Two normal distributions with equal spread. Probability of superiority = Φ(d/√2); share above the other group's average = Φ(d). At d = 0.5 the curves overlap by about 80%, and if you pick one person from each group at random the one from the higher group is ahead only about 64% of the time. Cohen suggested 0.2, 0.5 and 0.8 as small, medium and large, and he meant them as a last resort when you know nothing better about your field. A tiny effect can matter when it is applied to millions of people; a medium one can be useless for one person. Power is the chance a study detects an effect that is really there. It depends on the true effect, the sample size and the cut-off. Try it. Below, thousands of invented studies are run on the same true effect; significant ones are shaded. Interactive · Run 1,500 studies True effect d People per group Each study: two groups of n, normal data, two-sided t-test at p < .05, fixed random seed so the numbers repeat. Dark bars are significant studies. With d = 0.5 and 20 people per group, power is only about a third. With 64 per group it is about 80%, the usual target. Now notice the second number. When the true effect is small and the study is small, the few studies that reach significance are the ones that overshot: at d = 0.2 and 20 per group, power is about 10% and the "winners" average an effect around four times the truth. Journals that print only the winners print inflated effects. This is one reason replications so often come back smaller. P-hacking: the forking paths A p-value below .05 is meant to happen by luck one time in twenty when nothing is going on. That promise holds only if you decide what to test before you look. A researcher with ten outcome measures, four ways to exclude "bad" participants, and the freedom to stop when the result looks good has dozens of tests to choose from, and reports the one that worked. Simmons, Nelson and Simonsohn showed in 2011 that such ordinary flexibility could make almost anything "significant". Andrew Gelman called it the garden of forking paths: nobody has to cheat, because each fork seems reasonable at the time. Interactive · Nothing is going on. How often does something "work"? Many outcome measures Peek and stop early Outcomes measured 1,000 simulated studies per setting; both groups drawn from the same distribution, so every significant result is a false alarm. Outcomes: 20 per group, independent outcomes. Peeking: test after every 10 per group. With ten independent outcomes the chance that at least one is "significant" is 1 − 0.9510, about 40%. With twenty it is about 64%. Peeking at the data after every ten participants and stopping the moment p dips under .05 gives a false-alarm rate of about one in five (18% in the simulation) across ten looks. Real outcomes are correlated, so the real numbers are a little lower, but the lesson stands. The cure is not more careful people. It is committing to the plan first. The replication crisis, and the reforms that followed Around 2011 several things collided. Daryl Bem published nine experiments in a leading journal claiming evidence of precognition, using ordinary methods, which suggested ordinary methods could prove the impossible. The Dutch social psychologist Diederik Stapel was found to have fabricated data. And the false-positive paper appeared. Then people began to re-run studies. In 2015 the Open Science Collaboration, led by Brian Nosek, repeated 100 published experiments: 97 of the originals had significant results, while only about 36% of the replications did, and replication effects averaged about half the original size. Cognitive studies fared better than social ones. That does not say 64% of psychology is false, since a failed replication can itself be underpowered, and studies differ in how hard they are to repeat. It does say published effects are routinely inflated. The Many Labs projects asked a different question: do the same results appear across many sites? In the first, in 2014, 10 of 13 effects replicated consistently across 36 samples, one weakly, and two, flag priming and currency priming, did not. Test your own sense of what held up. Interactive · Held up, shrank, or failed? Guess, then reveal Guess each one. Your tally appears here. The reforms are practical. Preregistration means writing down hypotheses and the analysis plan before data arrive, which removes the forks. Registered reports go further: a journal reviews the question and method first and promises to publish whatever comes out, which removes the pull to print only winners; the Center for Open Science counts over 300 journals offering the format. Open data and materials let others check and reuse your work. Larger samples, shared multi-lab studies and reporting every effect size with its uncertainty are now common advice. Not everyone agrees on how far to go: some argue that strict rules suit confirmatory tests and fit badly with exploratory or qualitative work, and that fixing the incentives matters more than any checklist. Few dispute that the field is better checked than a decade ago. Who is in the study? WEIRD samples Most experiments recruit whoever is nearest: university students. In 2010 Joseph Henrich, Steven Heine and Ara Norenzayan reviewed the evidence in The Weirdest People in the World? and pointed out that the participants were overwhelmingly from societies that are Western, Educated, Industrialised, Rich and Democratic, a small slice of humanity, and that on several measures such people were outliers. The classic example is a visual illusion (the Müller-Lyer, in Sensation & Perception, 35/21), which fools people in industrial societies more than it fools people in some small-scale societies. Results about fairness games, moral reasoning and self-concept differ across cultures too. The remedy is plain: say who was studied, write the claim to match, recruit more widely, and be wary of "humans do X" from a campus sample. Some effects are universal, some are local, and without other samples you cannot tell which. What a graduate just sees A study is only as good as its ruler: a noisy or invalid measure hides real effects or measures something else. "Linked to" is not "caused by". Only random assignment removes a hidden third factor. Significant is not important. Ask how big the effect is, and how uncertain. Small studies are noisy, and the ones that reach print overshoot. Treat a lone small result as a hint. Flexibility in analysis quietly raises false alarms. The question is what was decided before the data. One result is a starting point; a result that repeats across labs is evidence. Who was studied limits who the finding is about. A failed replication is information, not an accusation. Where it shows up next Social Psychology (35/25): obedience, priming and group studies are where the replication story is most dramatic, and the tools above decide what you can still trust. Personality & Individual Differences (35/26): reliability and validity are the whole business of a test, and heritability claims depend on design. Econometrics (34/44) and Marketing Research and Analytics (34/59): the same logic of confounders, A/B tests and natural experiments in other disciplines. In the world: reading a news story about a "new study", judging a health claim, running a product experiment, or deciding whether a therapy has evidence. If you want to go deeper 📘 Textbook. Beth Morling, Research Methods in Psychology: Evaluating a World of Information (W. W. Norton, 5th edition). Widely used; built around judging claims in the news as well as designing studies. 📘 Textbook. Frederick Gravetter and Lori-Ann Forzano, Research Methods for the Behavioral Sciences (Cengage). The other standard syllabus text. 📘 Gentler. Stuart Ritchie, Science Fictions (2020). A readable account of fraud, bias, hype and the reforms. 🌐 Free text. Daniël Lakens, Improving Your Statistical Inferences: a free online book on effect sizes, power and open practices. 🌐 Free course. MIT OpenCourseWare 9.00SC Introduction to Psychology (Fall 2011, John Gabrieli), a full introductory course. The takeaway Psychology is hard to measure and easy to fool yourself about. The defences are the ones in this piece: measure reliably, assign at random when you can, report how big, run studies large enough to see the effect, and decide the analysis before you look. When a field applied them, a lot of famous results shrank or vanished, and that is the method working. Try this: the next time you read "a new study shows", find three things: how many people, whether they were assigned at random, and how big the effect was. Most headlines leave out at least two. Sources & Further Reading Sources. Reliability, validity, attenuation, Cohen's d and power are standard textbook methods (Morling; Cohen 1988). Open Science Collaboration, Estimating the reproducibility of psychological science, Science 349 (2015): 100 studies; 97 originals significant; about 36% of replications significant; effect sizes about half. Klein et al., Investigating variation in replicability, Social Psychology 45 (2014): 13 effects, 36 samples. Simmons, Nelson & Simonsohn, False-positive psychology, Psychological Science (2011). Hagger et al., multilab ego-depletion replication, Perspectives on Psychological Science (2016): 23 labs, d = 0.04, 95% CI −0.07 to 0.15. Ranehill et al. (2015) on power posing; Watts, Duncan & Quan (2018) on the marshmallow test. Henrich, Heine & Norenzayan, The weirdest people in the world?, Behavioral and Brain Sciences 33 (2010). Registered Reports: Center for Open Science (over 300 journals). All simulated numbers are invented illustrations computed live, and a one-person demo is not a study. 🌐 Lakens, Improving Your Statistical Inferences 🌐 MIT OCW 9.00SC Introduction to Psychology 🌐 Center for Open Science: Registered Reports 📘 Morling, Research Methods in Psychology Step 5. Sensation & Perception: The Brain's Best Guess ------------------------------------------------------ Why: How the brain builds a world from senses. You do not see what is in front of you. You see the brain's best guess about it, built from noisy signals in a few hundred milliseconds. The course teaches how the signals are made, how close the guess usually is, and what happens when it is wrong. .chips button{min-width:0} .row2{display:flex;gap:8px;flex-wrap:wrap} .row2 button{flex:1 1 90px;min-width:0} .sw{display:flex;gap:8px;margin:8px 0} .sw div{flex:1;height:54px;border-radius:10px;border:1px solid var(--line);font-family:-apple-system,sans-serif;font-size:.75rem;display:flex;align-items:flex-end;padding:4px 6px;color:#000;min-width:0} #cbSvg circle{cursor:pointer} Where is the hole in your vision? There is one in each eye, where the optic nerve leaves the retina and no receptors can detect light. At arm's length it is roughly the size of a tangerine. You never notice. The brain fills it in with the surroundings and gives you a seamless picture. That filling-in is the whole subject in miniature. The eye reports a flickering, blurred, upside-down image at the back of the eyeball; the brain delivers a steady, sharp, three-dimensional world with colours and objects. Between the two is inference. Someone who has taken a sensation and perception course never again thinks of seeing as a camera. The course in one box Who takes itPsychology majors, usually in year 2, plus neuroscience, optometry, audiology, design and human-factors students. The British Psychological Society's core areas include it under cognitive and biological psychology, and many US programmes list it as a core course. What it is forExplaining how energy from the world becomes experience, how well the system works, and why its mistakes are systematic. Comes afterBiological Psychology (35/20), for neurons and spikes. Leads toCognitive Psychology (35/22): attention, memory and thinking build on the same predictive machinery. What this piece gives youYou will not be able to design a vision test. You will see why you notice ratios not amounts, why detecting something is a decision, why three colour detectors make a million colours, and why you can look straight at a gorilla and miss it. Easier first reads: Color Explained (08/06), Perspective: How a Flat Surface Fakes Depth (08/07) and Light and Color: Why the World Looks the Way It Does (13/09). One-person demos here are not studies. Transduction: every sense starts by making spikes The senses differ in what they detect but share a first step, transduction: a receptor cell turns a physical event into an electrical change, and from there the message travels as spikes (see Biological Psychology, 35/20). Light bends a molecule in a photoreceptor; sound waves bend hair cells in the inner ear; pressure opens channels in a touch receptor; odour molecules bind to receptors in the nose. The brain only ever receives spikes, and learns what they mean. SenseReceptorWhat it detects VisionRods and cones in the retinaLight intensity and wavelength HearingHair cells in the cochleaAir-pressure waves TouchMechanoreceptors in skinPressure, vibration, stretch Smell and tasteChemoreceptorsMolecules BalanceHair cells in the inner earHead acceleration The first stage is astonishingly sensitive. In 1942 Selig Hecht, Shlaer and Pirenne had people sit in the dark and report flashes. Their results suggested that a rod cell can respond to a single photon, and that seeing a faint flash takes only about 5 to 14 photons being absorbed in the retina. The human eye has roughly 120 million rods, for dim light, and about 6 million cones, for detail and colour. Thresholds and Weber's law: noticing ratios How faint a light can you see? That is the absolute threshold. How much must something change before you notice? That is the just noticeable difference, the JND. Ernst Weber found in the 1830s that the JND is not a fixed amount. It is a fixed fraction of what you started with. Lift 100 grams and you notice an extra 2 g; hold a 1 kg bag and you need about 20 g more. Gustav Fechner built the field of psychophysics on that in 1860. ΔI = k × IWeber's law: the smallest noticeable change grows in proportion to the starting intensity. k (the Weber fraction) differs by sense. Interactive · The just noticeable difference Lifted weight Brightness Loudness Starting intensity Rounded textbook-style Weber fractions (lifted weight about 2%, brightness about 8%, loudness about 5%); real values depend on conditions and the person. The slider is logarithmic: 1 to 10,000 in arbitrary units. A 10 kg load needs about 200 g more before you feel it. Fechner turned the idea into a rule that sensation grows with the logarithm of the stimulus. Later S. S. Stevens argued that a power law fits many senses better, with the exponent differing by sense (about 0.6 for loudness and 3.5 for electric shock). Weber's law fails at the extremes of intensity, but it is a good first guess and explains why a candle is bright at night and invisible at noon. Interactive · Which bar is longer? Bar length Difference Top is longerBottom is longerReset score A one-person demo, not a study. Try the same percent difference at short and long lengths: Weber's law predicts similar accuracy; tiny lengths run into the screen's pixels. Signal detection: every "did I see it?" is a decision Weber and Fechner assumed a fixed threshold: below it you see nothing, above it you do. Real detection is noisier. A radiologist looks for a faint shadow, a lifeguard for a swimmer in trouble, a screener for a knife on a scan. The brain's internal signal fluctuates even when nothing is there, so on some nothing-trials it looks like something. Two things then decide what you say: how well the real signal stands out from the noise, and how willing you are to say "yes". Signal detection theory, developed in the 1950s and used in psychology by David Green and John Swets (1966), separates them. Sensitivity, d′, is the distance between the "noise" and "signal" distributions in standard-deviation units. The criterion is the internal cut-off for saying yes. Move the cut-off and you trade hits for false alarms without changing sensitivity at all. d′ = z(hit rate) − z(false-alarm rate)z is the normal-curve score; d′ = 0 is guessing, d′ near 3 is easy Interactive · Sensitivity and the cut-off Smoke alarm (say yes readily) Balanced Jury (say yes rarely) Sensitivity d′ Criterion (positive = more cautious) Equal-variance model: noise ~ N(0,1), signal ~ N(d′,1), "yes" when evidence exceeds d′/2 + criterion. With d′ = 1.7 and a balanced cut-off you get about 80% hits and 20% false alarms. Push the cut-off up and false alarms fall to 3%, but hits fall to 44%. The same observer with the same eyes is either a nervous or a lax judge, and a manager who punishes only misses will get more false alarms. The same model describes medical tests, eyewitness line-ups and spam filters. Colour vision: three detectors, millions of colours Colour is not in the light. Light has wavelengths; colour is what the brain makes of how three kinds of cone cells respond. Thomas Young proposed three receptors in 1802. We now know them as the S, M and L cones, tuned roughly to 420, 530 and 560 nm. Their sensitivities overlap heavily, so every light produces a triple of responses, and the brain compares them. A key consequence: different mixtures of light can give identical triples. Those are metamers, and you look at them on every screen: a display has only red, green and blue lights, yet it shows yellow, because red plus green light excites the L and M cones in the same proportion as pure yellow light. The cones cannot tell the difference. Interactive · Match a colour with two lights Target light (single wavelength) targetmix of two Cone sensitivities are rough bell curves peaking at 420, 534 and 564 nm, a teaching model rather than measured data. The mix uses 540 nm and 640 nm lights chosen to best fit the target. Screen colours are approximations, and your screen is itself making every colour from three primaries. In the model a mix of 540 nm and 640 nm light matches yellow at 580 nm exactly, but for blue-green at 500 nm no match exists, because you would need a negative amount of red. Real colour-matching experiments found the same, and in them an observer could match by adding red to the target instead. About one man in twelve and one woman in two hundred (of northern European ancestry; rates are lower elsewhere) have red-green deficiency, because the L and M cone pigments are alike or one is missing, so more lights become metamers to them. The brain's guesses: illusions and predictive perception The retinal image is ambiguous: the same image could come from a small near thing or a large far thing. The brain resolves it with assumptions built from experience. Hermann von Helmholtz called this unconscious inference in the 1860s. Modern versions, predictive processing, say perception is the brain's running prediction, corrected by incoming signals that report the error. Rao and Ballard proposed a model like this for visual cortex in 1999; Karl Friston, Andy Clark and Anil Seth have developed the wider case. Not everyone accepts that one principle explains the whole brain, and critics say the framework is too flexible to test, but its picture of perception as inference is widely used. Illusions are the clearest evidence, since they are what a sensible assumption does in an unusual case. Look at the two lines below: one has fins turned out, one turned in. Your task is to make them look equal. Interactive · Müller-Lyer: make them look equal Length of the lower line (as % of the upper) Equal length is 100%. Most people in industrial societies set the fins-in line a tenth or more longer than equal, but the size varies with the figure. One person on one screen is a demo, not data. When the lines are really equal (100%), they should look different. The leading explanation is that fins-out and fins-in look like the near and far corners of rooms and buildings, and size is judged accordingly. The illusion is weaker in some small-scale societies, which fits the idea that it depends on experience with carpentered corners, though explanations are debated. Henrich and colleagues used this example in their 2010 WEIRD paper (see Research Methods in Psychology, 35/19). Another demo is the blind spot itself. Interactive · Find your blind spot Gap between cross and dot Close your left eye. Look at the cross with your right eye only. Slowly move closer or further until the dot vanishes. The optic disc sits about 15 degrees from the point you look at, so the distance is about 3.7 times the gap. Another famous case is the 2015 photograph of "the dress", which people saw as blue-and-black or white-and-gold. Research suggests the difference comes from assumptions about the lighting, which the brain subtracts, so it is a guess made from different priors. Same signal, two sensible inferences. Attention: you see what you look for You cannot process everything, so attention selects. In 1999 Daniel Simons and Christopher Chabris had viewers count basketball passes in a video; about half of them failed to notice a person in a gorilla suit walking through the middle. This is inattentional blindness. A related effect, change blindness, was shown by Ronald Rensink and colleagues in 1997. Flick between two pictures with a brief blank in between and big changes go unseen for many seconds, because the blank hides the motion that would have attracted attention. Interactive · Spot the change With blank flash No blank (direct swap) New scene Tap the circle that changes colour. Times are measured on your device and depend on the scene; a demo, not an experiment. With the blank, most people take several seconds to find a single colour change. Without it, the change jumps out at once, because the swap makes a visible flicker. Magicians know this: the brain skips what is not attended, and what it skips it fills in with a guess. What a graduate just sees The brain gets spikes, not pictures. Seeing is inference from receptor signals. We notice ratios, not amounts: a change is judged relative to what was there. Detecting is deciding: sensitivity and criterion are different things, and a cautious judge is not a worse sensor. Colour is a three-number summary of light, so very different spectra can look identical. Illusions are good assumptions applied in unusual cases, and some depend on culture. There is a blind spot you never notice, filled in by the brain. What you do not attend to can be missed even in plain view. Be wary of "I saw it with my own eyes": the eyes are the first step of the inference. Where it shows up next Cognitive Psychology (35/22): attention and memory are the next stages; eyewitness error is perception plus memory. Biological Psychology (35/20): the cortex's visual and auditory areas are the lesion cases that taught us where perception is built. Consumer Behavior (34/58): packaging, shelf position and colour choices work through attention and the assumptions above. Machine learning (34/32): vision models learn their own features, and "adversarial examples" are machine illusions. In the world: airport screening, radiology, road safety (looked but failed to see), display and audio design, hearing aids, colour-blind-friendly charts. If you want to go deeper 📘 Textbook. Jeremy Wolfe and colleagues, Sensation & Perception (Oxford University Press / Sinauer, 7th edition, 2024). Widely assigned, clearly illustrated, with the dress discussion revised. 📘 Textbook. E. Bruce Goldstein and James Brockmole, Sensation and Perception (Cengage). The other standard syllabus text. 📘 Gentler. Anil Seth, Being You (2021): perception as "controlled hallucination", written for general readers. Contested in parts. 🌐 Free course. MIT OpenCourseWare 9.35 Sensation and Perception (Spring 2009). The takeaway Perception is an educated guess built from noisy signals. It notices ratios, treats every detection as a decision, summarises light in three numbers, and attends to only a slice of what arrives. Its mistakes are not random: they are the assumptions that usually serve us, shown at their limits. Try this: this week, find your blind spot and then watch an object appear to fill it. Or notice a change someone makes to a room or a screen that you only see once a friend points at it. Sources & Further Reading Sources. Transduction, Weber's law, signal detection and trichromacy are standard textbook material (Wolfe et al.; Goldstein & Brockmole). Weber (1834) and Fechner (1860); Stevens (1957) on power laws. Hecht, Shlaer & Pirenne, Journal of General Physiology (1942): about 5 to 14 photons absorbed. About 120 million rods and 6 million cones: standard textbook estimates. Green & Swets, Signal Detection Theory and Psychophysics (1966). Young (1802); cone peaks near 420, 530 and 560 nm. Colour deficiency rates: Colour Blind Awareness (about 1 in 12 men, 1 in 200 women, northern European ancestry). Helmholtz; Rao & Ballard, Nature Neuroscience (1999). Müller-Lyer cultural variation: Segall, Campbell & Herskovits (1966); Henrich, Heine & Norenzayan (2010). The dress: Lafer-Sousa, Hermann & Conway (2015); Brainard & Hurlbert (2015). Simons & Chabris, Perception (1999); Rensink, O'Regan & Clark, Psychological Science (1997). Wolfe et al., 7th edition (2024). All interactive numbers are computed live from simple models; weber fractions are rounded illustrative values. 🌐 MIT OCW 9.35 Sensation and Perception 📘 Wolfe et al., Sensation & Perception 📘 Seth, Being You Step 6. Cognitive Psychology: A Narrow Beam and a Rebuilding Crew ----------------------------------------------------------------- Why: Memory, attention and thinking. You attend to very little, hold about four chunks, and rebuild every memory as you use it. That design is clever, and it is why a confident witness can be sincerely wrong. You are sure you remember where you parked, who was at the party, which of two cars cut you off. You can see it. Then a photo, a receipt or another person disagrees, and the picture you were so sure of turns out to have been assembled, a little at a time, after the fact. Or: you are at a busy table, talking, and across the room someone says your name. You never heard the rest of that conversation, yet your own name pulled you out of this one. Or you try to keep a phone number in your head while a friend reads you another, and one of them simply vanishes. Cognitive psychology is the study of those ordinary moments: how a mind takes in a flood of input, keeps a little of it, stores a fraction of that, and rebuilds the rest on demand. A graduate stops trusting the feeling of "I clearly remember" and starts asking how the memory was built. This matters in classrooms, in cockpits, and, as we will see, in courtrooms. The course in one box Who takes itEvery psychology degree, usually in year 2. Neuroscience, linguistics, computer science (AI) and education majors take a version of it, and law students meet its memory half. What it is forExplaining how attention, memory, reasoning and language work, by experiments that time people, count their errors and compare conditions. Comes afterSensation & Perception (35/21), which itself needs Biological Psychology (35/20). Leads toDevelopmental Psychology (35/24), and beyond psychology to Behavioral Economics (34/46), Machine Learning (34/32) and, for lawyers, Civil Procedure & Evidence (35/08) and Criminal Law & Procedure (35/07). What this piece gives youYou will not be able to design an experiment, which is what Research Methods in Psychology (35/19) is for. You will have felt the limits of your own attention and memory on this page, and you will know why a confident eyewitness is not the same as an accurate one. Gentler first reads: Memory: How to Make Things Stick (17/04) and The Biases That Fool Everyone (17/02). The demos below are one-person toys timed on your device, not studies. Attention is a narrow beam, and most of the world falls outside it You are bombarded with far more than you can process, so the mind selects. In 1953 Colin Cherry played people two voices, one in each ear, and asked them to repeat one. They could, and could say almost nothing about the other voice, except that it was a voice, and that their own name, if it appeared there, often got through. Broadbent's 1958 filter model was the first answer: a bottleneck lets one channel through. Later models made the filter leakier, and the debate over how early or late the filter acts is still alive. The vivid modern version is inattentional blindness. In the famous 1999 "gorilla" experiment by Daniel Simons and Christopher Chabris, viewers counting basketball passes often failed to notice a person in a gorilla suit walking through the scene; in most groups about half missed it. Nothing is wrong with their eyes. What you see is what you attend to, and you cannot tell what you are missing, because the missing is not marked. The cleanest way to feel the limit is the Stroop effect (J. Ridley Stroop, 1935). Reading a word is so automatic that it intrudes when you try to name the colour of its ink. The conflict costs time you can measure. Interactive · Stroop on your own device Tap the colour of the ink, not the word. Six easy trials (word and ink agree), then six hard ones (they conflict). Your device times each tap. Ready? Start Press Start. A twelve-trial toy shows the effect for you. It is not a measurement of you, and timing on a phone screen adds noise. Most people are slower on the conflicting trials, often by a tenth of a second or more, and make more errors. Nothing "loaded" more information; two automatic processes simply fought over one response. A graduate carries the lesson everywhere: attention is a single narrow beam that can be split only at a cost, which is why a phone call hurts driving even when the phone is hands-free. Working memory holds about four chunks, not seven digits Hold a phone number in your head and you are using working memory: the scratchpad where you keep things while you use them. George Miller's 1956 paper "The Magical Number Seven, Plus or Minus Two" made 7 famous. Later work, above all Nelson Cowan's 2001 review, argued the true limit is nearer four chunks once you stop people from rehearsing and regrouping. The word that matters is chunk: a unit that is one thing to you. The letters F-B-I-C-I-A-N-A-S-A are ten items, or three chunks if you know the acronyms. Baddeley and Hitch (1974) described this scratchpad as a small set of parts, a loop for sounds, a sketchpad for pictures and a controller that directs attention. A graduate remembers the practical points. The store is small. It is easily overwritten. And the way out is not more effort but better chunking, which is why experts seem to have enormous memory in their own field, and ordinary memory outside it. Interactive · Digit span, with and without chunks One at a time In groups of three Digits to remember Ready? Show digits Check Pick a length and press Show digits. Digits are generated fresh each time, shown for 0.7 s each (groups get a longer pause between them). Try 7 plain, then 9 in threes. Most readers manage around six or seven plain digits and do noticeably better on longer strings when they are grouped, because three groups of three are three chunks. That is not a trick. It is the whole mechanism of memory improvement: the number stays small; the chunk gets bigger. Encoding and retrieval: what goes in decides what comes out Remembering has two halves. Encoding is how the information is processed on the way in; retrieval is how you get it back. In 1972 Fergus Craik and Robert Lockhart argued that deeper processing (what does this mean to me?) beats shallow processing (what does it look like?), and the result has held up well. A second idea, encoding specificity, says a cue works to the extent it matches how the memory was laid down. In a well-known 1975 study, Godden and Baddeley had divers learn word lists on land or underwater; words were recalled better in the place they were learned. The context effect is real but modest and not always found, so treat "study where you'll be tested" as a small bonus, not a rule. And then there is time. Hermann Ebbinghaus (1885) measured his own forgetting of nonsense syllables and found that loss is fastest at first and then flattens. A 2015 replication of his work (Murre & Dros) found the same shape. The graduate's practical conclusion is the spacing effect: reviewing something just as it begins to fade slows the next fall. The toy below uses a simple exponential model with made-up constants, so it shows the shape of the idea, not your forgetting. Interactive · Forgetting curve and review (illustrative model) No review One review, day 1 Days 1, 3, 7, 14 Model: retention = e−t/S. Start with S = 3 days (made up). Each review resets retention to 100% and multiplies S by 2.5 (also made up). Real forgetting depends on the material and the person. With no review, almost nothing is left by day 30 in this model (under 0.01%). With one review, about 2%. With four spaced reviews, about 87%. The numbers are invented; the reason spacing beats cramming is the finding. Memory is rebuilt each time, not replayed The folk picture is a video recording. The research picture, which goes back to Frederic Bartlett's 1932 Remembering, is a reconstruction: when you recall, you assemble the memory from fragments plus what you know about how things usually go. That makes memory efficient and error-prone, and the errors are not random. They move toward what makes sense. A neat way to see it is the paradigm that James Deese (1959) and Henry Roediger and Kathleen McDermott (1995) turned into a standard tool. Study a list of words that all circle one unlisted word, for example bed, rest, tired, dream, nap, then later say which words you saw. People often "recognise" the unlisted word, here sleep, nearly as readily as words that were really on the list. Try it. Interactive · A false memory, on demand Fifteen words appear, 1.2 s each. Then twelve words follow, one at a time; tap whether each was on the list. Ready? Start the list Was on the listWas not Press Start. The list is from the Roediger–McDermott (1995) study, built around the unlisted word "sleep". In the test, half the words were studied, two were unrelated, and two were related but not shown. The same machinery explains Elizabeth Loftus's best-known results. In 1974 Loftus and John Palmer showed people films of car crashes and then asked how fast the cars were going when they "smashed into" or "contacted" each other. The "smashed" group guessed about 41 mph; the "contacted" group about 32. A week later, those asked with "smashed" were more likely to report broken glass that was never in the film. In 1995 Loftus and Jacqueline Pickrell planted a false childhood memory of being lost in a shopping mall, and about a quarter of the 24 participants reported remembering it. Critics have noted that "remembering" in these studies is partial, and that planting a whole memory is harder than planting a detail. The settled point is narrower and still important: suggestion after the event can change what people report, and they can't feel the change. Eyewitnesses: sincere, confident and sometimes wrong This is where memory science meets the law. In the United States, the Innocence Project reports eyewitness misidentification in about 69% of the first 367 DNA exonerations it counts (the share was about 75% in older snapshots, when there were fewer cases). Countries differ in how they run and weigh identifications, but human memory is the same everywhere. The American National Academy of Sciences' 2014 report Identifying the Culprit reviewed the evidence and recommended practices such as blind lineup administration and recording how confident a witness is at the first identification. The fairest summary has two halves, and researchers disagree about their weights. Many studies show that confidence at trial, after months of repeated questioning, tells you little about accuracy. John Wixted and Gary Wells (2017) argued that the first confidence statement, from an uncontaminated lineup, is quite informative: high-confidence initial picks are usually right, and low-confidence picks often wrong. Both can be true, because "confident" at trial is not the same thing as "confident at the start". To see the structure, borrow signal detection theory. A witness gets a familiarity signal for each face. The culprit's signal is stronger by an amount d′ (sensitivity). The witness also has a bar (the criterion): how familiar a face must feel before they pick it. A lineup puts the possible culprit among five look-alike fillers. When the culprit is absent, a witness who feels pressure to choose will still choose. Interactive · A lineup, simulated 20,000 times Witness sensitivity d′ Bar for picking someone Toy model: six faces, equal-variance normal familiarity, the witness picks the most familiar face if it clears the bar. Culprit present and absent are equally likely. A real witness is messier. Notice what the sliders show. With d′ at 0, the witness knows nothing and the picks are noise. With a low bar, the witness picks someone nearly every time, and in culprit-absent lineups that someone is innocent. Raising the bar costs correct IDs but removes most of the false ones. That is why reformers focus on procedure: blind administration, fair fillers, and not telling the witness "good, you picked the suspect". Heuristics: fast rules that mostly work, and sometimes don't Amos Tversky and Daniel Kahneman's 1974 paper "Judgment under Uncertainty: Heuristics and Biases" showed that people judge probability by shortcuts. Availability: what comes easily to mind seems common, so after a news story people overestimate plane crashes. Representativeness: what resembles a type seems likely to belong to it, which is how the "Linda" problem (1983) leads many to rate "feminist bank teller" as more probable than "bank teller", though a subset can never be more probable than the whole. Anchoring: a number you've just heard drags your estimate toward it. Gerd Gigerenzer and others argue that heuristics are not just errors: in the right environment, simple rules do well, and the "biases" are partly artefacts of how problems are posed. Kahneman's side says the errors are systematic and costly. Both agree that the way you present numbers matters. A classic case from Tversky and Kahneman involves a witness again. Interactive · The taxi problem A cab hit a pedestrian at night. Some cabs in the city are Blue, the rest Green. A witness says the cab was Blue. Tested in similar light, the witness gets the colour right some of the time. Share of cabs that are Blue Witness accuracy Illustrative numbers; the 15% and 80% are the classic textbook version. Imagine 1,000 cabs and count. With 15% Blue cabs and an 80% accurate witness, the chance the cab really was Blue is about 41%, not 80%. Out of 1,000 cabs, 150 are Blue and the witness calls 120 of them Blue; of 850 Green cabs, the witness wrongly calls 170 Blue. So 120 of 290 "Blue" reports are right. In Tversky and Kahneman's report, most people answered near the witness's 80%, ignoring the base rate. Counting people instead of multiplying probabilities (Gigerenzer's "natural frequencies") makes the right answer much easier to find. Language and thought: does the word shape the idea? Edward Sapir and Benjamin Lee Whorf are credited with the idea that language shapes thought. The strong version, that you cannot think what you cannot say, has been rejected: people think about things for which they have no word, and babies think before they speak. The weak version, that language nudges attention and memory, has support. Berlin and Kay (1969) showed that colour terms across languages follow a near-universal order, suggesting biology constrains the vocabulary. And Winawer and colleagues (2007) found that Russian speakers, whose language has separate basic words for light blue (goluboy) and dark blue (siniy), were quicker to tell apart two blues that straddle that line, an advantage that vanished when they had to repeat a string of digits at the same time. Other claims, such as grammatical gender making speakers describe objects differently, are more contested, and several famous language-and-thought results have been hard to replicate. Say "language nudges" and stop there. The replication crisis touches this whole course. In 2015 the Open Science Collaboration tried to repeat 100 published psychology studies; about 97% of the originals had significant results and about 36% of the replications did. Cognitive-psychology effects, which are usually within-person and measured in milliseconds, fared better than social ones. Stroop, serial-position and false-memory effects replicate easily. Treat very large claims from small studies with caution, and see Research Methods in Psychology (35/19). What a graduate just sees Attention is a narrow beam: what you did not attend to is not "seen but forgotten", it was never stored. Working memory holds about four chunks. Better chunking, not harder trying, is how people hold more. How deeply you process something when it goes in decides how easily it comes out. Review at the edge of forgetting beats re-reading in one sitting. Recall is reconstruction. A vivid, detailed, confident memory can still be wrong. Questions asked after an event can change the memory without any feeling of change. A witness's accuracy depends on procedure: how the lineup was run and how soon confidence was recorded. When someone gives you a probability, ask for the base rate; count cases instead of multiplying fractions. Language nudges thought; it does not cage it. Where it shows up next Developmental Psychology (35/24): memory, attention and reasoning are not fixed. Children's working memory and theory of mind grow, and older adults' processing speed slows while vocabulary keeps growing. Behavioral Economics (34/46): the heuristics-and-biases programme is where much of the field began; its replication record is mixed, as the economics piece says. Machine Learning (34/32): human memory as a rebuild-from-cues system resembles models that fill gaps with likely content, which is both a strength and the source of confident errors. Law: Criminal Law & Procedure (35/07) and Civil Procedure & Evidence (35/08) meet eyewitness memory directly. Courts in different countries weigh identifications differently, and a number of jurisdictions have adopted lineup reforms. In the world: air traffic and cockpit design (attention and workload), interface design, teaching (spacing and retrieval practice), police interviewing, and the warnings on any "I remember it clearly" claim. If you want to go deeper 📘 Textbook. Michael Eysenck and Mark Keane, Cognitive Psychology: A Student's Handbook (Routledge / Psychology Press). A common UK and international choice, thorough and research-heavy. 📘 Textbook. Daniel Reisberg, Cognition: Exploring the Science of the Mind (W. W. Norton). A common US choice, easier on the first read. 📘 Gentler. Daniel Schacter, The Seven Sins of Memory (Houghton Mifflin, 2001), and Daniel Kahneman, Thinking, Fast and Slow (2011), noting that some priming studies in the latter did not replicate. Elizabeth Loftus's TED talk "The fiction of memory" is short. 🌐 Free course. MIT OpenCourseWare 9.00SC Introduction to Psychology (Fall 2011), whose lectures include memory and cognition; the free Noba chapters cover each topic here. The takeaway The mind is a narrow beam of attention, a four-chunk scratchpad and a long store that it rebuilds each time it is used. That design is efficient, and it means vividness and confidence are not evidence of accuracy. Good procedure, such as spaced review, blind lineups and counting cases instead of trusting a feeling, is how people protect themselves from their own minds. Try this: this week, when you are sure about a detail from last month, find something that can check it (a message, a photo, a receipt). Notice how often the gist is right and a detail has quietly moved. Sources & Further Reading Sources. Cherry 1953 and Broadbent 1958 (selective attention); Stroop 1935; Simons & Chabris 1999, Perception 28; Miller 1956, Psychological Review 63; Cowan 2001, Behavioral and Brain Sciences 24; Baddeley & Hitch 1974; Craik & Lockhart 1972; Godden & Baddeley 1975; Ebbinghaus 1885 and Murre & Dros 2015, PLoS ONE (replication); Bartlett 1932; Deese 1959 and Roediger & McDermott 1995, JEP: Learning, Memory, and Cognition 21; Loftus & Palmer 1974, Journal of Verbal Learning and Verbal Behavior 13 (mean estimates by verb: "smashed" about 41 mph, "contacted" about 32 mph); Loftus & Pickrell 1995, Psychiatric Annals 25 (24 participants, about a quarter reported the false event); Innocence Project (eyewitness misidentification in 69% of 367 DNA exonerations in its current count, earlier figure about 75%); National Research Council 2014, Identifying the Culprit; Wixted & Wells 2017, Psychological Science in the Public Interest 18; Tversky & Kahneman 1974, Science 185, and 1983 (Linda); Gigerenzer & Hoffrage 1995; Berlin & Kay 1969; Winawer et al. 2007, PNAS 104; Open Science Collaboration 2015, Science 349. Every number an interactive shows was computed in node: taxi problem 120/290 = 41.4%; lineup model at d′ 1.5, bar 1.5 gives 45% correct picks, 35% no pick when the culprit is present and a 5.5% innocent-suspect pick when absent; forgetting model, day 30: under 0.01% with no review, 2.1% with one review, 87.2% with reviews on days 1, 3, 7 and 14 (S grows from 3 to about 117 days). 🌐 MIT OCW 9.00SC Introduction to Psychology 🌐 Elizabeth Loftus, "The fiction of memory" (TED) 📘 Eysenck & Keane, Cognitive Psychology: A Student's Handbook 📘 Schacter, The Seven Sins of Memory This describes research, not medical or legal advice. If you are struggling with your memory, mood or wellbeing, a doctor or local service can help; findahelpline.com lists free helplines by country. Step 7. Developmental Psychology: How Minds Are Built, and What Later Research Revised -------------------------------------------------------------------------------------- Why: How minds grow from infancy to old age. From Piaget's stages to false beliefs, attachment and the long rebuild of adolescence. What held up, what shifted, and why heritability is not a pie chart. A two-year-old hides her face behind her hands and announces that you can't see her. A four-year-old watches you look for your keys in the wrong drawer and, quietly, knows you are looking in the wrong place. A sixteen-year-old does something spectacularly risky when friends are watching and something sensible when alone. A seventy-year-old forgets a name but explains the situation better than anyone else in the room. These are not separate mysteries. They are four points on one long road, and developmental psychology is the study of the road: what changes, in what order, why, and how much it depends on the child, the family and the century. A graduate of this course stops asking "is it nature or nurture?" and starts asking "which differences, in which population, at which age, under which conditions?". The course in one box Who takes itEvery psychology degree, usually in year 2; education, nursing, social work and paediatrics students take a version. Programmes vary on whether it covers the whole lifespan or mainly childhood. What it is forDescribing how thinking, feeling and relating change from infancy to old age, and testing the explanations. Comes afterCognitive Psychology (35/22) and Learning & Behavior (35/23). Leads toSocial Psychology (35/25), Personality & Individual Differences (35/26) and Clinical Psychology (35/27), and on to teaching, paediatrics and policy. What this piece gives youYou will not be able to assess a child. You will know the big theories, what has been revised, and why children in different places are not interchangeable. Gentler first reads: Attachment: The Blueprint (02/01), How Children Think (02/07), Why Teenagers Aren't Broken (02/03) and Nature, Nurture and You (02/04). This goes further and shows the evidence and its critics. The demos are toys, not studies. Piaget: children think differently, and later research revised the timetable Jean Piaget (1896 to 1980) watched his own children and Swiss schoolchildren and concluded that they are not small adults with less knowledge. They think in a different way, and the way changes in steps. His four stages: sensorimotor (birth to about 2, knowing through action), preoperational (about 2 to 7, symbols and pretend but thinking tied to appearances), concrete operational (about 7 to 11, logic about real things) and formal operational (from about 11 or 12, abstract and hypothetical thought). Two ideas have survived best: children actively build understanding rather than absorbing it, and they can be systematically wrong in ways that make sense from the inside. His best-known test is conservation. Pour water from a short wide glass into a tall thin one and a five-year-old says there is now more. Piaget read this as a lack of the logic of reversibility. Later researchers argued he underestimated young children, because his tasks were hard to follow. In a 1974 study, James McGarrigle and Margaret Donaldson had a "naughty teddy" mess up the arrangement by accident rather than the experimenter doing it on purpose, and many more children then gave the right answer. In the same spirit, Martin Hughes's 1975 police-doll task showed that children as young as three and a half could work out where to hide a doll from a policeman, a task Piaget's "three mountains" test had said they could not do. At the other end, researchers used looking time to ask what babies know. In the 1980s Renée Baillargeon and colleagues reported that infants as young as about three and a half to five months looked longer at an "impossible" event, such as a screen rotating through a hidden box, than at a possible one. They concluded that babies have early knowledge about objects. That claim is contested. Critics point out that looking longer may reflect novelty rather than surprise, that search tasks (reaching for a hidden toy) show later competence, and that some pupil-dilation and replication attempts, especially at 10 months, have not found the expected effect. The honest summary: infants have more early structure than Piaget thought, how much is still argued, and the stage boundaries are softer than the textbook diagram. Interactive · How a looking-time experiment is built (toy model) Tracks hidden objects Does not Does not, but the "impossible" display looks different Familiarisation trials All constants are made up to show the logic. Looking time = 2 + 14·e−0.4n s on trial n. Test events: a possible one (the same as before) and an impossible one, which adds 6 s for an infant who tracks hidden objects, or for one who only notices that it looks different. The toy shows why the debate is hard. Habituation shortens looking at a repeated event; a test that recovers looking is read as "this is surprising". But two different infants, one who understands objects and one who merely finds a new display interesting, can produce the same pattern. Good studies are designed to tell them apart. Vygotsky: the mind grows between people Lev Vygotsky (1896 to 1934) died of tuberculosis at 37, and most of the English-speaking world met him decades later, through Mind in Society (1978). Where Piaget's child is a lone scientist, Vygotsky's child learns thinking tools by using them with others: language first used between people becomes private speech and then inner thought. His best-known idea is the zone of proximal development: the gap between what a child can do alone and what they can do with a little help. Teaching works best inside that gap, not at the level already mastered nor far beyond reach. Wood, Bruner and Ross (1976) called the adult's changing support scaffolding: giving more help when the child struggles and less as they succeed. Vygotsky's work is more a framework than a set of precise predictions, so it has been harder to test than Piaget's. It still shapes teaching practice across the world, from tutoring to the way many early-years programmes use guided play. The cross-cultural point: in many communities learning happens by children joining in adult work rather than in school-style lessons, a pattern Barbara Rogoff and others have documented. Attachment: the first relationships John Bowlby, a British psychiatrist, argued in the 1950s and 60s that infants are built to seek closeness to a caregiver, and that this bond is not just about food. Harry Harlow's 1950s monkey experiments showed baby monkeys preferred a soft cloth surrogate to a wire one that held the milk. Mary Ainsworth, who had observed mothers and babies in Uganda, designed the Strange Situation (1970s): a short lab procedure in which a one-year-old sees a parent leave and return. Babies were sorted by their reaction at reunion: secure (upset perhaps, but comforted and soon back to play), avoidant (little visible distress, avoiding the parent on return), resistant or ambivalent (clingy and angry), and later disorganised. In many studies, most babies are classed secure. But the rest of the picture differs by place. A 1988 meta-analysis by van IJzendoorn and Kroonenberg reported that the patterns shifted across countries; studies in northern Germany found more avoidant babies than American norms, and a study in Sapporo, Japan found more resistant ones. The disagreement is over what that means. One reading is that cultures raise babies differently (a German emphasis on early independence, Japanese babies rarely separated from their mothers, so the procedure stresses them more). Another is that the test, built in the US and the UK, does not measure the same thing everywhere. Both are taken seriously. The strongest evidence about attachment and early care comes from the Romanian orphanages of the 1980s. The Bucharest Early Intervention Project (Nelson, Zeanah, Fox and colleagues, 2007) randomly assigned institutionalised children to foster care or to staying put; those moved earlier, especially before about two, did better on many measures. The English and Romanian Adoptees study (Michael Rutter and others) found that longer deprivation went with longer-lasting difficulties. The lesson is "sensitive period", not "critical window that slams shut": the earlier the good care, the better the recovery, and recovery is still possible later. Another caution: attachment style correlates with later outcomes, but a child's own temperament and the many things that change in life make the link far from destiny. Theory of mind: knowing that others know different things A theory of mind is the understanding that other people have beliefs, desires and knowledge that may differ from yours and from reality. The classic test is the false-belief task (Wimmer and Perner, 1983; the Sally–Anne version by Baron-Cohen, Leslie and Frith, 1985). Try it. Interactive · Predict, then reveal Next sceneStart over In the basketIn the box Press Next scene. Adults answer "the basket" without a thought: Sally believes what she last saw. Many three-year-olds say "the box", because that is where the marble really is. Most children pass around four to five. Wellman, Cross and Watson's 2001 meta-analysis of 178 studies found the same shift from below chance to above chance across the many countries studied, though the age and order of steps differ: in a 2011 study (Shahaeian and colleagues) children in Iran and Australia learned some steps in different orders, an effect linked to how families talk about knowledge. Some experiments report that babies as young as 15 months look longer at scenes that violate a false belief (Onishi and Baillargeon, 2005), but replication attempts have been mixed, and whether this reveals early understanding or something simpler is debated. It is also worth saying that theory-of-mind findings in autistic people are contested: the long-standing claim that autistic people lack a theory of mind has been challenged by researchers who describe a two-way difficulty in understanding across different minds, not a one-way deficit. Adolescence and ageing: change that is neither disorder nor decay Adolescence is a long, normal rebuild. A widely cited 2018 paper (Sawyer and colleagues) argued that it now stretches from about 10 to 24, because puberty comes earlier and adult roles come later. The brain is reorganising: connections are pruned, and the prefrontal regions that help with planning and self-control keep maturing into the twenties. But "the brain isn't mature until 25" is a slogan, not a threshold; there's no age at which a switch flips, and the exact timing varies by region and by person. One well-replicated finding is about when teenagers take risks. In a driving-game study (Gardner and Steinberg, 2005), adolescents took more risks when peers were watching than when alone, while adults did not change nearly as much. The usual reading, Laurence Steinberg's "dual-systems" account, is that a reward system matures before the control system. Critics argue the picture is more nuanced, since adolescents can reason well in calm conditions and risk-taking can also be exploration and learning. For more, see Why Teenagers Aren't Broken (02/03). At the other end, ageing is not simply decline. Psychologists separate fluid abilities (quick reasoning, processing speed, which tend to fall from the twenties) from crystallised ones (vocabulary and knowledge, which keep growing). In 2015, Joshua Hartshorne and Laura Germine analysed large online samples and found that abilities peak at very different ages: processing speed around the late teens and early twenties, working memory in the late twenties to early thirties, and vocabulary much later, into the sixties or seventies. Their result is cross-sectional and from one research programme, so cohort differences blur it. Dementia is a disease, not normal ageing. Interactive · Who is at their peak at this age? Age The bars show approximate peak ranges as summarised from Hartshorne and Germine (2015): speed 18 to 22, working memory 27 to 33, vocabulary 60 to 70. They are simplified ranges, not a curve, and the exact ages vary by task and generation. Nature and nurture, done properly "Is it genes or environment?" is a badly formed question, a bit like asking whether the area of a field comes from its length or its width. The better question is: in this population, how much of the difference between people goes with differences in their genes? That share is the heritability, and it has three properties people forget. It describes a population, not a person; "50% heritable" does not mean half of your height is from genes. It depends on the environment: it is higher where everyone has similar conditions, lower where conditions differ. And it is not fate: a trait can be strongly heritable and still change a lot when the environment changes, as the PKU metabolic disorder does with diet, or as average height has done over the last century (NCD-RisC's 2016 analysis found a gain of about 20 cm for South Korean women born between 1896 and 1996). The numbers are real, though. A 2015 meta-analysis of fifty years of twin studies (Polderman and colleagues, 17,804 traits, about 14.5 million twin pairs) put the average heritability across traits at about 49%, with shared family environment usually small once genes are counted. Eric Turkheimer summarised the pattern as three laws: all human behavioural traits are heritable; the effect of being raised in the same family is smaller than the effect of genes; and a substantial part of the variation is not explained by genes or families. Critics of twin methods argue about their assumptions, and measured DNA often explains less than twin studies imply (the "missing heritability" puzzle). Interactive · Same genes, different heritability Similar environments In between Very different environments Genetic differences (SD) Shared family environment (SD) Individual environment and chance (SD) Simulation of 3,000 identical-twin pairs and 3,000 fraternal pairs (identical twins share all genes, fraternal share half on average). The estimate is Falconer's: heritability = 2 × (identical correlation − fraternal correlation). A model, with the usual twin assumptions. Slide the environment from "similar" to "very different" while leaving the genetic differences untouched. Heritability falls from about 0.9 to about 0.2. The genes did not weaken; the environment simply began to explain more of the gaps. Whose children? The WEIRD problem and cross-cultural findings In 2010 Joseph Henrich, Steven Heine and Ara Norenzayan pointed out that most behavioural science samples come from people who are Western, Educated, Industrialised, Rich and Democratic (WEIRD), a small and unusual slice of humanity. In development this matters a lot. Sleeping alone, school-style questioning, "talking to" babies, and independence as the goal are cultural choices, not universals. Attachment categories vary, theory-of-mind steps vary in order, and the famous "marshmallow test", once thought to predict adult success, was shown in a larger, more diverse sample (Watts, Duncan and Quan, 2018) to be far weaker once family background was accounted for. What does appear widely is the basic outline: infants everywhere form attachments, children everywhere pass false-belief tasks in the preschool years, and adolescents everywhere seek peers. The details are cultural. What a graduate just sees Children are not small adults; their errors have an inside logic, and a task's wording can hide what they know. Timetables of "stages" are rough. Ages for when children can do things move when the test is made fairer. Learning happens in the gap between what a child can do alone and with help; scaffolding means help that fades. A bond with a caregiver is a base for exploring, and the earlier poor care is repaired, the better. Passing the false-belief task means seeing that someone can hold a belief you know is wrong. Adolescent risk-taking depends on the audience, and the brain's rebuild has no fixed finish line. Different abilities peak at different ages; knowledge grows when speed falls. Heritability describes differences in a population and changes with the environment; it never says what one person will become. Ask who was in the sample before generalising to "children". Where it shows up next Social Psychology (35/25): conformity and group identity peak in adolescence, and theory of mind is the base of every social judgement. Personality & Individual Differences (35/26): temperament, the heritability toolkit (twins, adoption) and the stability of traits from childhood on. Clinical Psychology (35/27): attachment-informed therapy, how early adversity affects later health, and how conditions such as autism and ADHD are understood across the lifespan. Education and policy: the early-years evidence behind parental leave, quality childcare and preschool programmes, and the debate over how to teach (guided play, direct instruction, scaffolding). In the world: paediatric care, social work, adoption and foster-care practice, school design and the ageing-population debate. If you want to go deeper 📘 Textbook. Laura Berk, Child Development or Development Through the Lifespan (Pearson). A standard, clear text widely assigned in the US and beyond. 📘 Textbook. Robert Siegler and colleagues, How Children Develop (Worth / Macmillan), strong on cognition and on the evidence behind each claim. 📘 Gentler. Alison Gopnik, The Philosophical Baby (2009), and Margaret Donaldson, Children's Minds (1978), the classic that challenged Piaget with the "naughty teddy" and police-doll tasks. 🌐 Free course. MIT OpenCourseWare 9.85 Infant and Early Childhood Cognition (Fall 2012), and the free Noba developmental chapters. The takeaway Minds are built in steps, but the steps are softer than the diagrams, and they depend on help from others and on the culture around them. Early care matters and later repair is possible. Genes and environment do not split the credit like a pie: heritability is a statement about differences within a population, and it moves when the environment does. Try this: this week, watch a child (or remember being one) solve a problem with a little help. Notice what help worked, and how it faded as they got it. That is a zone of proximal development at work. Sources & Further Reading Sources. Piaget's stages and conservation (standard textbook accounts); McGarrigle & Donaldson 1974, Cognition 3; Hughes 1975 (doctoral thesis, Edinburgh) as reported in Donaldson 1978; Baillargeon, Spelke & Wasserman 1985; replication doubts: Pätzold & Liszkowski 2020 (PLoS ONE, pupillometry, 10-month-olds); Vygotsky 1978 (Mind in Society); Wood, Bruner & Ross 1976; Bowlby (Attachment and Loss, 1969 on); Harlow 1958; Ainsworth et al. 1978; van IJzendoorn & Kroonenberg 1988, Child Development 59; Nelson et al. 2007, Science 318; Rutter and the English and Romanian Adoptees study; Wimmer & Perner 1983; Baron-Cohen, Leslie & Frith 1985; Wellman, Cross & Watson 2001, Child Development 72 (178 studies); Shahaeian et al. 2011, Developmental Psychology 47; Onishi & Baillargeon 2005; Sawyer et al. 2018, Lancet Child & Adolescent Health 2; Gardner & Steinberg 2005, Developmental Psychology 41; Hartshorne & Germine 2015, Psychological Science 26; Polderman et al. 2015, Nature Genetics 47 (17,804 traits, about 14.5 million twin pairs, average heritability 49%); Turkheimer 2000, Current Directions in Psychological Science 9; NCD Risc 2016, eLife 5:e13410 (20.2 cm for South Korean women, 1896 to 1996 birth cohorts); Henrich, Heine & Norenzayan 2010, Behavioral and Brain Sciences 33; Watts, Duncan & Quan 2018, Psychological Science 29. Interactive numbers computed in node: twin simulation with 3,000 pairs each, genetic SD 1: similar environments (individual SD 0.3) gives heritability estimate 0.89 against a true 0.92; very different (SD 2) 0.19 against 0.20; the middle preset 0.44 against 0.44. 🌐 MIT OCW 9.85 Infant and Early Childhood Cognition 🌐 Noba Project 📘 Berk, Child Development 📘 Gopnik, The Philosophical Baby This describes research, not medical or parenting advice. If you are worried about a child's development, talk to a paediatrician or local service; if you are struggling yourself, findahelpline.com lists free helplines by country. Step 8. Social Psychology: How Other People Change What We Do ------------------------------------------------------------- Why: How other people change what we do. The classic experiments say that ordinary people go along, obey and look away. The re-checked versions say something more careful, and more interesting: the situation matters a great deal, and so does what people think the situation is. In 1964 a young woman was murdered in Queens, New York, and a newspaper told the country that thirty-seven neighbours watched and did nothing. The story became the opening page of nearly every social psychology textbook. It has also, in the decades since, turned out to be wrong in most of the details that made it shocking. That arc, a vivid story, a famous experiment, a tidy lesson, then a slow correction, is a fair summary of the whole subject. Social psychology asks how the presence of other people changes what one person thinks, says and does. Its best ideas survive. Many of its best-known demonstrations were less clean than the textbooks said. A graduate of this course keeps both: the ideas, and the habit of asking how a classic study was actually done. The course in one box Who takes itPsychology majors, usually in year 2; it is one of the core areas in the British Psychological Society's accreditation list and in the American Psychological Association's guidelines for the undergraduate major. Sociology, business and law students often take it as an elective. What it is forExplaining how groups, authority, first impressions and beliefs about other people shape behaviour, and how to tell a solid finding from a famous one. Comes afterResearch Methods in Psychology (35/19), which is where you learn why a single study proves little. Leads toOrganizational Behavior (34/53) and Consumer Behavior (34/58), and it sits beside Behavioral Economics (34/46) and Developmental Psychology (35/24). What this piece gives youYou will not be able to run an experiment. You will know six classic findings, what each actually showed, what its critics say, and how well it has held up. Easier first reads: The Biases That Fool Everyone (17/02) and Why You Understand Less Than You Think (17/10). Conformity: Asch and the line that was obviously wrong Solomon Asch's question was simple: will people state something they can plainly see is false, just because a group does? In his experiments (reported from 1951), a volunteer sat with several others, all but one of them secretly working with the experimenter, and judged which of three lines matched a standard line. The answer was obvious. On 12 of the 18 trials the confederates, one after another, gave the same wrong answer, and the real participant answered near the end. Alone, people made fewer than 1% errors. In the group, about 36% of answers on the critical trials went along with the wrong majority. About three quarters of participants conformed at least once, and about a quarter never did. That last number is the one textbooks tend to leave out: most answers were still correct, and a survey of textbooks in 2015 found that only one of twenty major texts reported that most responses defied the majority. The size of the group matters up to a point. A single dissenting-looking companion barely moves people; two or three make a real difference; beyond about three or four the effect levels off. The tool below draws that curve and lets you run a toy group of fifty simulated volunteers. Interactive · How many confederates does it take? People in the room giving the wrong answer Run 50 simulated volunteers Curve is approximate and rounded, following the commonly reported shape of Asch's group-size results; secondary sources differ by a point or two. The 50 volunteers are a toy model with a made-up spread of suggestibility (some people never conform), not data. Without JavaScript: with three or more people giving the wrong answer, about a third of answers conformed in Asch's experiments; with one, only a few per cent. Why did they conform? Asch interviewed them. Some doubted their own eyes, some knew the group was wrong but did not want to stand out, and a few came to see the lines the group's way. Psychologists later split this into informational influence (the group probably knows something) and normative influence (disagreeing has a social cost). John Turner and colleagues argued that the data fit a third idea better: we follow those we see as "people like us". Conformity also varies by place and time. A 1996 meta-analysis by Bond and Smith, covering 133 studies in 17 countries, found more conformity in more collectivist societies and a decline in the United States over the decades after the 1950s. Obedience: Milgram and the shock machine In 1961–62 at Yale, Stanley Milgram told volunteers they were helping a study of learning. A "teacher" read word pairs to a "learner" in the next room and gave an electric shock for each mistake, rising in 15-volt steps to 450 V. The learner, an actor, protested and eventually fell silent. If the teacher hesitated, the experimenter in a grey coat gave a fixed series of prods, the first being "Please continue." In the best-known condition, 26 of 40 men (65%) went all the way to 450 V. The number was shocking, and so was Milgram's reading of it: ordinary people will harm a stranger when an authority tells them to. But Milgram ran many variations, and they are the real finding. Obedience was not a trait of the people; it moved with the situation, from nearly everyone to hardly anyone. Interactive · One experiment, many results Percentages who went to the maximum 450 V, as reported for Milgram's variations (about 40 participants each). Telephone instructions are often given as 20.5% or 21%. Without JavaScript: baseline 65%; learner in the same room 40%; learner's hand forced onto the plate 30%; instructions by telephone about 21%; run-down office 47.5%; two fellow "teachers" refuse 10%; a fellow "teacher" who delivers all the shocks 92.5%. The critics are serious. Gina Perry, who went through Milgram's archive for her 2012 book Behind the Shock Machine, argued that the published account did not match what she found, including that only about half of participants fully believed the shocks were real, and that two thirds of those who did believe disobeyed. Alex Haslam and Stephen Reicher argue that people did not simply obey: they went on when the prod appealed to science ("The experiment requires you to continue") and stopped when it sounded like a bare order. Their reading is "engaged followership": participants identified with the study's goal, not with the learner. A 2009 partial replication by Jerry Burger, which stopped at 150 V for ethical reasons, found rates close to Milgram's at that point. The shock is real; the "people are robots" reading is not. Attribution: why we blame the person and excuse ourselves You see a driver cut across three lanes. You think: what a jerk. When you do it, the exit was suddenly there and the mirror was blocked. Attribution theory, started by Fritz Heider in the 1950s, studies how we explain behaviour: by something in the person (a disposition) or something in the situation. The consistent human habit is to over-read the person. Lee Ross named it the fundamental attribution error in 1977. The classic demonstrations: in 1967 Edward Jones and Victor Harris had people read an essay praising or attacking Fidel Castro, and told them the writer had been assigned the side. Readers still guessed the writer leaned that way. In Ross, Amabile and Steinmetz's 1977 quiz game, students were assigned by coin toss to write hard questions or to answer them. The questioners looked cleverer to everyone, including the answerers, though the role had made it so. Harold Kelley's 1967 covariation model gives a tidy rule for reasoning well. Ask three questions about the behaviour: does everyone do it (consensus), does this person do it to everything (distinctiveness), does this person do it every time (consistency)? Interactive · Sam laughs at a comedian. Why? Consensus: do other people laugh at this comedian? Yes, most doNo, hardly anyone Distinctiveness: does Sam laugh at other comedians too? No, only this oneYes, at nearly all Consistency: does Sam laugh at this comedian every time? Yes, every timeNo, only tonight Kelley's three textbook patterns are shown; the other five combinations are ambiguous and the model says so. Later studies (McArthur, 1972) found that people tend to under-use the consensus information. Without JavaScript: low consensus, low distinctiveness and high consistency points to the person; all three high points to the comedian; low consensus, high distinctiveness and low consistency points to the circumstances of tonight. Culture matters here too. In a well-known study Michael Morris and Kaiping Peng (1994) found that American and Chinese participants explained the same animated fish scene differently, Americans leaning on the fish's inner traits and Chinese participants on the group around it. It fits the larger point that much of social psychology was built on university students in a few rich countries, which Joseph Henrich, Steven Heine and Ara Norenzayan called WEIRD samples (Western, Educated, Industrialised, Rich, Democratic) in 2010. Attitudes and persuasion: when behaviour changes belief Common sense says attitudes drive behaviour. A good deal of the time it runs the other way, and Leon Festinger's cognitive dissonance theory (1957) says why: holding two clashing thoughts is uncomfortable, and the cheapest fix is often to change the belief, not the act. The famous test (Festinger and Carlsmith, 1959) had 71 male Stanford students do dull peg-turning, then asked some to tell the next person it was fun, for either $1 or $20. Which group came to like the task more? Interactive · Predict, then reveal Afterwards, who said the boring task had really been enjoyable? The $1 groupThe $20 groupNeither: no difference Pick one, then read the result. Without JavaScript: the $1 group rated the task as more enjoyable than the $20 group, the opposite of what simple reward would predict. The $1 group did, and the $20 group only slightly more than people who were never asked to lie. Dissonance theory says $1 is too little to justify the lie, so the mind adjusts the belief; $20 is justification enough. The theory has had rivals (Daryl Bem's self-perception account) and refinements, and the "free-choice" variant, in which people supposedly like what they chose more, has been criticised for a statistical flaw. But the core pattern, that acts shape beliefs especially when we feel responsible for them, is one of the more durable ideas in the field. On persuasion itself, Richard Petty and John Cacioppo's elaboration likelihood model (1986) says there are two routes: a central one, where people weigh arguments when they are motivated and able, and a peripheral one, where cues such as a confident speaker or a famous face do the work. Strong arguments matter most when people are paying attention. When they are not, packaging does. And attitudes predict behaviour only loosely: Richard LaPiere's 1934 trip with a Chinese couple found almost every hotel and restaurant served them, while most of the same businesses later said by mail that they would not. What people say and what they do can differ. Groups and prejudice: how little it takes to make "us" In the early 1970s Henri Tajfel divided schoolboys in Bristol into two groups by a trivial rule (they were told it was a preference for paintings by Klee or by Kandinsky) and then asked each boy to hand out points to anonymous members of the two groups. The boys knew none of them and would gain nothing. Yet they favoured their own group, and often chose options that gave their group fewer points in total as long as the gap over the other group got bigger. This minimal group paradigm underlies social identity theory: part of who we are comes from the groups we belong to, and we want our group to look good against others. Interactive · Hand out the points You are in the "blue" group. Choose one column: points for an anonymous blue member (top) and an anonymous "green" member (bottom). Choose a pair. An invented matrix in the spirit of Tajfel's, not his actual numbers. Real participants were schoolboys, and what they picked on matrices designed to pull the strategies apart is what the study reported. Without JavaScript: in the matrix, giving 13 points to both groups gives each 13; giving 7 to your group and 1 to the other leaves a gap of 6 but costs your group 6 points. Muzafer Sherif's Robbers Cave study (Oklahoma, 1954) is the famous other half: twenty-two boys at a summer camp, split into two groups, competed for prizes and soon despised each other, then made peace when forced to cooperate on shared goals. The conclusion that competition breeds hostility and common goals reduce it is widely taught. The critique is that the story is tidier than the record: earlier versions of the study in 1949 and 1953 did not produce the same result (in one, the boys turned on the experimenters), and Michael Billig has argued that the staff were effectively a third group steering events. What has held up better is Gordon Allport's 1954 contact hypothesis: contact between groups reduces prejudice under conditions such as equal status and shared goals, supported by a large meta-analysis of hundreds of studies by Thomas Pettigrew and Linda Tropp. Some newer popular ideas deserve caution. The Implicit Association Test is widely used to measure unconscious bias, but its scores are only modestly stable from one session to the next and predict individual behaviour weakly, and changing a score has not reliably changed behaviour. It may tell you about group-level patterns better than it tells you about a person. Bystanders: the story and the evidence The Kitty Genovese story, as taught for decades, said thirty-eight people watched a murder from their windows and did nothing, out of apathy. In 2007 Rachael Manning, Mark Levine and Alan Collins went back to the record: far fewer people had a clear view, many who heard noise did not realise what was happening, and at least one person called the police. In 2016 the New York Times itself called its original reporting flawed and said it had grossly exaggerated the number of witnesses. The better evidence is laboratory. In 1968 John Darley and Bibb Latané had students talk by intercom, one at a time, with others in separate rooms; one of the "others" seemed to have a seizure. Students who believed they were the only listener reported it before the fit ended 85% of the time. With one other possible listener the figure was 62%, and with four others, 31%. Latané and Darley called the mechanism diffusion of responsibility: with others around, each person feels less personally responsible. Interactive · One person, or the whole room? The 85, 62 and 31 are from Darley and Latané (1968), the proportion reporting before the fit ended. The "someone helps" row is arithmetic that assumes each listener acts independently, which real groups, who also watch each other, will not. Without JavaScript: each person is less likely to act in a group (85%, 62%, 31%), but if five listeners each had a 31% chance, the chance that at least one acts is about 84%, close to the lone person's 85%. Look at what the arithmetic says. The effect is real for each individual, but the group as a whole may still produce help about as often as one person would. That is why later work added caveats. A 2011 meta-analysis by Peter Fischer and colleagues found the bystander effect strong in non-dangerous emergencies and weaker in dangerous ones, and weaker when bystanders are not strangers. A 2020 study by Richard Philpot and colleagues of more than 200 real CCTV recordings of street conflicts in the UK, the Netherlands and South Africa found that in over 90% someone intervened, and more bystanders meant more chance of help. The bystander effect is real, but people are not generally apathetic. What held up The Open Science Collaboration (2015) tried to repeat 100 psychology studies; about 36% of the repeats were statistically significant against 97% of the originals, and the repeats showed effects about half the size. Social psychology fared worse than cognitive psychology (about a quarter replicated against about a half). Treat the list below as a rough report card, not a verdict on any one study. Held up well:solid Asch-style conformity in its broad direction; anchoring-like context effects; the contact hypothesis; the bystander effect in non-dangerous settings. Held, but smaller or messier:shrank Milgram's obedience (the rate depends on setup, and the interpretation changed); cognitive dissonance in its simplest forms; the fundamental attribution error (strong in the lab, culture-bound in size). Failed or collapsed:failed Many social-priming effects, such as words about old age making people walk slower; power posing changing hormones; ego depletion. See Research Methods in Psychology (35/19). Disputed:disputed The Stanford Prison Experiment, which lasted six days in 1971, said guards turned cruel by role. Thibault Le Texier (2018) reported from the archive that the guards were coached on how to behave, so it is better read as a demonstration of what people do when told what is expected than as evidence of a natural tendency. What a graduate just sees Most people in an Asch room still gave the right answer; the effect is real and also smaller than the legend. Obedience swings from 10% to 90% depending on the setup, so ask what the situation is, not what kind of person someone is. When judging someone else, check the situation first; when judging yourself, you already do. Acts change beliefs: if you were paid little for something you did, you will find reasons it was worth it. A trivial label is enough to start "us and them"; shared goals are better at undoing it. In a crowd each person feels less responsible, but a named person asked directly usually helps. A famous study is a lead, not a fact: look for the replications, the sample, the critics. Where it shows up next Organizational Behavior (34/53): group decisions, conformity in meetings, authority and safety culture. Consumer Behavior (34/58) and Behavioral Economics (34/46): social proof, defaults, norms, and which of the "nudges" are weaker than advertised. Clinical Psychology (35/27): stigma, group therapy and attribution styles in depression. In the world: jury deliberation, online pile-ons, workplace safety, military ethics, and how emergency courses teach "you, in the red coat, call 999". If you want to go deeper 📘 Textbook. Elliot Aronson, Timothy Wilson and Robin Akert, Social Psychology (Pearson, many editions), or David Myers, Social Psychology (McGraw-Hill). Both are standard undergraduate texts; check for a recent edition that covers replication. 📘 Gentler. Gina Perry, Behind the Shock Machine (2012) on Milgram, and Gina Perry, The Lost Boys (2018) on Robbers Cave. 🌐 Free text. The free Noba psychology textbooks have a module on Conformity and Obedience, written for introductory courses. 🌐 Free course. MIT OpenCourseWare 9.00SC Introduction to Psychology (Fall 2011) is the free introductory course. The takeaway Other people shape what we say, do and believe, and the situation can matter more than character. But the best-known demonstrations are more qualified than their reputations: most people resisted in Asch's rooms, Milgram's results depended on the setup, the Genovese crowd was exaggerated, and several famous effects shrank on re-testing. Try this: the next time someone's behaviour annoys you, write down two situational explanations before you settle on a trait. Sources & Further Reading Sources. Asch, "Opinions and Social Pressure," Scientific American 1955, and the 1951 monograph (about 36% of critical responses conforming; about 74% at least once; control error under 1%); a 2015 survey of twenty textbooks on Asch (and an earlier one in 1990); Turner on referent informational influence; Bond and Smith, Psychological Bulletin 1996. Milgram, Journal of Abnormal and Social Psychology 1963 and Obedience to Authority (1974): variations as summarised in Wikipedia, "Milgram experiment"; Perry 2012; Haslam, Reicher and Birney on engaged followership; Burger, American Psychologist 2009. Jones and Harris 1967; Ross 1977; Ross, Amabile and Steinmetz 1977; Kelley 1967; McArthur 1972; Morris and Peng 1994; Henrich, Heine and Norenzayan, Behavioral and Brain Sciences 2010. Festinger and Carlsmith, 1959 (71 male Stanford students); Petty and Cacioppo 1986; LaPiere 1934. Tajfel, Billig, Bundy and Flament, "Social categorization and intergroup behaviour," European Journal of Social Psychology 1971; Sherif et al., Intergroup Conflict and Cooperation: The Robbers Cave Experiment (1961); Billig; Allport, The Nature of Prejudice (1954); Pettigrew and Tropp, 2006. Darley and Latané, Journal of Personality and Social Psychology 8 (1968), 377–383; Manning, Levine and Collins, American Psychologist 2007; the New York Times acknowledgement of 2016; Fischer et al., Psychological Bulletin 2011; Philpot et al., American Psychologist 2020. Open Science Collaboration, Science 2015; Le Texier, American Psychologist 2019. Calculations on this page (the 84% and 85% group figures, the toy simulation) were run in node. This piece describes research and is not a guide to any individual's behaviour. 📘 Perry, Behind the Shock Machine 📘 Aronson, Wilson and Akert, Social Psychology 🌐 Noba Project Step 9. Personality & Individual Differences: What Varies Between People ------------------------------------------------------------------------ Why: Personality, intelligence and how much is inherited. Why people differ, how psychologists measure it, how stable it is, and how much of it is inherited. Most of what you were taught about this in a magazine quiz is the weakest part. Two siblings grow up in the same house, eat the same food, go to the same school. One is calm, tidy and quiet; the other is loud, late and fearless. A teacher can usually predict who will volunteer for the school play. A parent, asked how much of that difference was the house and how much was something else, often finds the answer is neither of the two things they expected. Most of psychology looks for what people share: how memory works, how vision works. Differential psychology, the study of individual differences, asks what varies between people and how far. It has the most useful measurements in psychology, and some of the most abused. This piece covers how traits are measured, how stable they are over a life, what an IQ score is, what "heritable" does and does not mean, and where the field's popular claims fall apart. The course in one box Who takes itPsychology majors, usually in year 3. The British Psychological Society's accreditation core lists individual differences as a separate area, and most American programmes teach a personality course and an intelligence unit. Education, HR and genetics students often take it too. What it is forDescribing how people differ in stable ways, measuring that fairly, and not over-reading it. Comes afterResearch Methods in Psychology (35/19): you need correlation and reliability from there. Leads toClinical Psychology (35/27), and later Organizational Behavior (34/53) and Consumer Behavior (34/58). What this piece gives youYou will not be able to score a test. You will know what a trait is, why a short quiz is a weak measurement, why "40% heritable" does not mean "40% genetic", and why a famous finding may belong to one culture. Easier first reads: Nature, Nurture and You (02/04) and The Biases That Fool Everyone (17/02). The Big Five, and what a test is If personality is real, it should be in the dictionary. That is the lexical hypothesis: the differences that matter to people get words. In 1936 Gordon Allport and Henry Odbert combed an English dictionary for about 18,000 terms describing people. Decades of work with factor analysis, which finds the few dimensions behind many correlated ratings, converged on five broad ones: openness, conscientiousness, extraversion, agreeableness and neuroticism (emotional instability), the "Big Five" or OCEAN. A "trait" here is a tendency measured on a continuum, with most people in the middle, not a type. A test is a handful of questions. One question is a poor measure: "I am the life of the party" depends on the party. Add questions about the same trait and the random noise averages out. The formula is the Spearman–Brown prophecy: if a single item has reliability r, a test of k items has reliability k·r ÷ (1 + (k − 1)·r). reliability = k·r ÷ (1 + (k − 1)·r)r is the reliability of one item; k is the number of items Interactive · Why one question is not a measurement Reliability of a single item Number of items on the scale The formula is exact; the single-item values are illustrative (real items vary). A common rule of thumb asks for about 0.7 or more for research and higher for decisions about individuals. Without JavaScript: with a single-item reliability of 0.3, ten items give about 0.81 and two items give about 0.46. Now try ten items yourself. These come from the International Personality Item Pool (IPIP), a public-domain bank of items that Lewis Goldberg started so that researchers would not have to pay for tests. Answer how well each describes you. This is for illustration, not a measurement: two items per trait is far too few, and no score here means anything about you. Interactive · Ten IPIP items (for illustration) Show my five numbersClear Answer at least one pair, then press the button. Nothing is stored or sent. Answers: 1 = very inaccurate, 5 = very accurate; reversed items are flipped before averaging. Notice what the output looks like: five dimensions, each with a position on a scale. A good adult measure has about 10 to 20 items per trait, and the trait scores correlate with real outcomes in the way you would expect, but only modestly. Conscientiousness predicts job performance and health to a small-to-moderate degree; extraversion predicts who is sociable at parties. Compare the Myers–Briggs Type Indicator, the best-known workplace "test". It sorts people into 16 types by cutting four continuous scales at the middle, so two people very close to the cut get different types; studies of retesting find that many people get a different type on a second try after some weeks. Its dimensions overlap with four of the Big Five, but the type labels throw information away, and it is not generally used in research. HEXACO, a competing model, adds a sixth factor, honesty–humility. Stability and change: the same person, different person There are two senses of "does personality change?" Rank-order stability asks whether someone who is more extraverted than most at 20 is still more extraverted than most at 40. Mean-level change asks whether people on average become more extraverted. They can differ completely: everyone can grow calmer and keep their ranking. The classic meta-analysis (Brent Roberts and Wendy DelVecchio, 2000) pooled 152 longitudinal studies and found test–retest correlations of about 0.31 in childhood, 0.54 in college, 0.64 at 30, and then up to about 0.74 for those aged 50 to 70. Later work (Bleidorn and colleagues, 2022) questions the late plateau and finds little further rise after about 25; the late-life data were thin. The direction of the finding is sturdy: personality is far from fixed in childhood, and gets more settled but never frozen. Interactive · How much do people keep their place? Simulated people: 400 draws with the given correlation (Roberts and DelVecchio 2000). The scatter is computed here, not real data. Intervals are several years, and the correlation is a summary, not a rule for any one person. Without JavaScript: with a correlation of 0.31 about 42% of people stay in the same third of the group over time; at 0.74 about 61% do. On average, people become more conscientious, agreeable and emotionally stable from their twenties into middle age: the maturity principle (Roberts, Walton and Viechtbauer, 2006). It is an average, of course, and some people move the other way. Major life changes, such as a first job or a long relationship, seem to matter more than time alone, but the causal story is not settled. Intelligence and its tests In 1904 Charles Spearman noticed that children who did well on one school subject tended to do well on the others and named the shared factor g. Today the strongest cognitive-ability models (the Cattell–Horn–Carroll model) arrange abilities in layers, with g at the top and narrower skills below it, such as verbal reasoning, spatial ability and processing speed. Alfred Binet built the first practical test in 1905 to find French schoolchildren who needed extra help; it was not meant to rank people as fixed. Modern IQ scores are not "points of cleverness". They are a place on a bell curve, scaled so that the population average is 100 and the standard deviation is 15 (some older tests used 16). A score of 130 is two standard deviations up, about the 98th percentile. And every score has error: even good tests have a standard error of measurement of a few points. Interactive · An IQ score is a range Score reported Standard error of measurement SD 15SD 16 Percentiles use the normal curve; the 95% range is score ± 1.96 × the standard error. Real error varies by test and by score. Without JavaScript: a score of 100 is the 50th percentile; with a standard error of 3 the 95% range is about 94 to 106; a score of 130 is about the 98th percentile on a scale with an SD of 15. IQ scores predict school grades and, less strongly, job performance and some health outcomes, which is why they are among the most useful numbers in psychology and also why misuse is dangerous. Early twentieth-century IQ testing was tied to eugenics and used to justify restrictions on immigration and forced sterilisation in the United States, and group differences in scores have been misread as proof of innate differences. Howard Gardner's "multiple intelligences" are popular in classrooms, but they are mostly not supported by tests, which find that his abilities correlate positively, and the "learning styles" idea has not held up in experiments. Scores have also been rising. The Flynn effect is the rise in average test scores of about three IQ points per decade through the twentieth century in many countries. In Norway, Denmark and elsewhere it has levelled off or reversed since the 1990s, and a 2018 study by Bratsberg and Rogeberg found the reversal appears within families, which points to the environment (schooling, media, testing habits) and not to a change in genes. Heritability: what the number does and does not say "Intelligence is about 50% heritable." What does that mean? Heritability is the share of the variation in a population, in one setting, that is associated with genetic differences. It is not the share of one person's trait that comes from genes, it says nothing about how changeable the trait is, and it depends on how much the environment varies. The classic picture (Richard Lewontin's seeds): plant genetically varied seeds in uniform, good soil and nearly all the difference in height is genetic; plant the same seeds in patchy soil and much of the difference is environmental. The genes did not change, so the heritability did. Twin studies estimate it by comparing identical twins (who share all their genes) with fraternal twins (about half). Falconer's formula: heritability h² = 2(rMZ − rDZ), where the r's are the twin correlations. The shared-family environment is c² = 2rDZ − rMZ, and the rest is non-shared environment (plus measurement error), e² = 1 − rMZ. Interactive · Twins, then soil Identical twins' correlation (MZ) Fraternal twins' correlation (DZ) Variety of environments (the soil) Presets are illustrative, not measured values. Falconer's estimate is a rough tool: it assumes no gene–environment interaction and equal environments for both kinds of twin. In the soil part, genetic variance is fixed at 100 and h² = 100 ÷ (100 + environmental variance). Without JavaScript: identical twins correlate 0.8 and fraternal 0.5 give h² = 0.6, c² = 0.2, e² = 0.2. With genetic variance 100 and environmental variance 100, h² = 0.5; with 400, h² = 0.2. What do real studies find? In twin studies of IQ, heritability tends to be low in childhood and to rise toward adulthood, in the region of 40% in childhood to well over half in adults, a pattern that surprises people who expect genes to fade with age. Eric Turkheimer's "three laws" of behaviour genetics (2000) summarise a vast literature: all human traits are partly heritable; family environment shared by siblings matters less than expected; a large share of variation is not explained by genes or by shared family at all. Turkheimer's own 2003 finding that IQ heritability was lower in poor US families is a good example of how contested the details are: a 2016 meta-analysis (Tucker-Drob and Bates) found it in US studies but not in the western Europe or Australian data. Molecular genetics adds a puzzle. When researchers look directly at DNA, the traits that twin studies call highly heritable are linked to thousands of tiny genetic differences, and the total they capture is smaller than the twin estimate (the "missing heritability"). Polygenic scores built from these differences predict less within a family, between siblings, than across unrelated people, partly because they pick up environment and family background too. Culture: whose people are being measured? Most psychology experiments are done on university students, in a few rich countries. A 2008 estimate by Jeffrey Arnett found that about 96% of participants in leading psychology journals came from countries with about 12% of the world's population, and in 2010 Joseph Henrich, Steven Heine and Ara Norenzayan named the problem: WEIRD samples (Western, Educated, Industrialised, Rich, Democratic), which are, on several measures, unusual people. The Big Five does appear in many languages and countries, which is a strength. But the picture is less tidy at the edges: a study of the Tsimane forager-farmers in Bolivia (Gurven and colleagues, 2013) did not recover the five clean factors; some lexical studies find slightly different structures; and the sex differences in personality scores are, oddly, larger in some richer and more gender-egalitarian countries (Schmitt and colleagues, 2008), a result researchers explain in competing ways. Culture shapes even perception. The Müller-Lyer illusion, two equal lines that look unequal when one ends in outward-pointing fins and the other in inward-pointing ones, was tested in many societies by Marshall Segall and colleagues in the 1960s. Henrich and colleagues used it as an example of a basic effect that varies: people raised in carpentered, right-angled environments tend to be more fooled than some people from small-scale societies. Try it. A one-person demo tells you nothing about a culture, but it shows the illusion is real. Interactive · Make the two lines look equal Length of the lower line, as % of the upper The upper line is always 200 units long. Most people who set the lower line to look equal choose a length under 100%, because the outward fins make the line look longer. One person on one screen is a demo, not a study. Without JavaScript: the two lines are exactly equal, but the one with the outward fins looks longer to most viewers. Oversold: what personality science did not show Several popular ideas about individual differences are weaker than their reputation. Left-brain and right-brain "types" are not supported; a study of over 1,000 brain scans (Nielsen and colleagues, 2013) found no evidence that people use one hemisphere more. "We use only 10% of our brains" is false. The "Mozart effect", a brief boost in spatial reasoning after listening to Mozart reported with 36 students in 1993, did not hold up beyond a small arousal effect in later meta-analyses. Learning styles, the idea that people learn better in the style they prefer, has failed in experiments (a 2008 review by Pashler and colleagues found no adequate evidence). The Dunning–Kruger effect is a good case. The popular version says "the less you know, the more confident you are". The 1999 paper by David Dunning and Justin Kruger found that people in the lowest quartile on tests overestimated their rank, while people in the top quartile slightly underestimated it, but nobody in the bottom group rated themselves above the top group. Critics (for example Gignac and Zajenkowski, 2020) argue that much of the pattern is a statistical artefact: if self-estimates are noisy at all, they will pull toward the middle, so low scorers seem to overestimate and high scorers to underestimate. Dunning and others reply that real miscalibration remains. The tool below shows how much the pattern arises from noise alone. Interactive · A Dunning–Kruger graph from noise How well people really know their standing (0 = not at all) 1,000 simulated people with a made-up level of insight; no "blind spot" is built in. Bars show the average actual percentile (top bar of each pair) and the average self-estimated percentile (second) for each quartile. Without JavaScript: with an insight correlation of 0.5, the lowest quartile averages about 13 in reality and about 31 in self-estimate; the top quartile about 88 and about 69. What a graduate just sees A trait is a position on a scale, not a box; a "type" quiz throws information away. One question is not a measurement; reliability grows with the number of items, with diminishing returns. People keep their rank better with age, but they never freeze. An IQ score is a range on a bell curve, and 130 and 120 are not very different people. Heritability describes a population in one environment; it is not the share of a person that is genetic, nor a limit on change. A big, sure-sounding number from a twin study is a starting point, not an answer. Ask who was in the sample before trusting that a finding is about "people". When a result is surprising and tidy (Mozart, left-brain types, Dunning–Kruger), check the statistics first. Where it shows up next Clinical Psychology (35/27): personality disorders, how traits shade into disorder, and why diagnosis is a cut-off on a scale. Organizational Behavior (34/53): hiring tests, what they predict and which ones are mostly folklore. Developmental Psychology (35/24): temperament in infants and how it changes. Statistics (34/05) and Research Methods in Psychology (35/19): reliability, correlation and sampling are the tools used here. In the world: school testing, hiring, the genetics of education, and the headlines that quote a polygenic score. If you want to go deeper 📘 Textbook. Randy Larsen and David Buss, Personality Psychology: Domains of Knowledge About Human Nature (McGraw-Hill, many editions); David Funder, The Personality Puzzle (Norton, many editions). 📘 Gentler. Stuart Ritchie, Intelligence: All That Matters (Hodder, 2015) and Kathryn Paige Harden, The Genetic Lottery (Princeton, 2021) on heritability and what it means. 🌐 Free text. The Noba chapters, for example Personality Traits, and the International Personality Item Pool itself. 🌐 Free course. Open Yale Courses PSYC 110, Introduction to Psychology (Paul Bloom), with lectures on personality, intelligence and mental illness. The takeaway People differ in stable, measurable ways, mostly along five broad dimensions, and measurement works when many items are averaged. Traits settle with age and never freeze; IQ is a range on a curve; heritability is a fact about a population in one environment, not a verdict on any one person. Be most sceptical of the tidy, surprising claims. Try this: next time you read "X% heritable" or "scientists found a gene for Y", ask three questions: in what population, in what environment, and does it say anything about changing it? Sources & Further Reading Sources. Allport and Odbert, Trait-Names: A Psycho-Lexical Study (1936); Goldberg's IPIP (ipip.ori.org; the items used are from the IPIP 50-item Big-Five markers); Spearman–Brown formula (Spearman 1910, Brown 1910); Roberts and DelVecchio, Psychological Bulletin 2000 (152 longitudinal studies; .31, .54, .64, about .74); Bleidorn and colleagues, 2022, on the late-life plateau; Roberts, Walton and Viechtbauer, Psychological Bulletin 2006; Spearman 1904; Binet and Simon 1905; Carroll 1993; Flynn effect of about 3 points per decade (Trahan et al. 2014 give 2.93) and the Norwegian reversal (Bratsberg and Rogeberg, PNAS 2018); Falconer's formula; Lewontin 1970s on heritability; Turkheimer, Current Directions in Psychological Science 2000 and Psychological Science 2003; Tucker-Drob and Bates, Psychological Science 2016 (14 studies, 24,926 twin and sibling pairs; the effect appeared in US studies, not in Western Europe or Australia); Arnett, American Psychologist 2008; Henrich, Heine and Norenzayan 2010; Segall, Campbell and Herskovits (1966); Gurven et al. 2013; Schmitt et al. 2008; Nielsen et al. 2013; Pashler et al. 2008; Rauscher et al. 1993; Dunning and Kruger 1999; Gignac and Zajenkowski 2020. Computed in node from the code on this page: reliability at r = 0.3 for k = 10 is 0.811 and for k = 2 is 0.462; a score of 130 is the 97.7th percentile at SD 15. The Myers–Briggs claim is stated generally; this piece does not give a retest percentage. This piece describes research and is not a personal assessment. 📘 Funder, The Personality Puzzle 🌐 International Personality Item Pool 🌐 Open Yale Courses PSYC 110 Step 10. Clinical Psychology: Where Normal Ends and Care Begins --------------------------------------------------------------- Why: Mental disorders and the therapies that work. What makes something a disorder, how the two big manuals disagree, why causes are never just one thing, which treatments have good evidence, and why most of the world's people who need help never get it. This piece describes how clinical psychology works. It is not medical advice, and nothing here can tell you whether you or someone you know has a condition. If you are struggling, a doctor or a local mental health service is the right place to start; findahelpline.com lists free helplines by country. Everyone has had a week when they could not sleep, felt flat, or worried so much that they could not think. Almost nobody thinks that week is an illness. Yet there are people for whom that week lasts a year, and who cannot work, eat or be with their family. Somewhere between these lies a line, and it is not a line that a blood test can draw. That is the central puzzle of clinical psychology: how to tell ordinary suffering from a disorder, how to describe what is wrong without pretending to know more about the cause than we do, and how to help. A graduate of this course keeps three habits: treat diagnoses as useful descriptions with fuzzy edges, treat causes as many things at once, and ask of every treatment how well it was tested. The course in one box Who takes itPsychology majors, usually in year 3; it is the gateway to clinical training (a doctorate or a master's degree plus supervised practice in most countries, the route and the title differ by country). Nursing, social work and medical students take versions of it. What it is forDescribing, understanding and treating mental disorders, and knowing how strong the evidence is. Comes afterBiological Psychology (35/20) and Personality & Individual Differences (35/26). Leads toClinical and counselling training; it also feeds Organizational Behavior (34/53) and, with Developmental Psychology (35/24), the study of how problems begin. What this piece gives youYou will not be able to diagnose anyone, and this piece does not offer a self-test. You will know how a disorder is defined, what the main conditions are, what works and how well, and what the global picture looks like. Easier first reads: How to Actually Calm Down (01/03), Stress vs Burnout (01/05) and Aging Well (02/10). What makes something a disorder The definition in the American Psychiatric Association's manual runs roughly like this: a mental disorder is a clinically significant disturbance in thinking, emotion regulation or behaviour that reflects a problem in psychological, biological or developmental processes, and that usually comes with distress or disability. It adds exclusions: an expected response to a loss is not a disorder, and neither is behaviour that is merely socially deviant, or in conflict with society, unless it reflects a dysfunction in the person. Philosophers have argued about this for decades. Jerome Wakefield's "harmful dysfunction" idea says a disorder needs both a failure of some mental mechanism to do what it evolved to do and harm to the person. Critics ask who decides what a mechanism is for. In practice, clinicians use three working questions: how much does it distress the person, how much does it get in the way of life, and how far is it outside what the person's culture expects? The history shows how much the line moves. Homosexuality was listed as a mental disorder in early editions of the American manual; the American Psychiatric Association's board voted to remove it in 1973, and the World Health Organization removed it from the international list in 1990. In 1973 David Rosenhan published "On Being Sane in Insane Places", reporting that eight people who feigned hearing a voice were admitted to psychiatric hospitals; the paper was influential, and Susannah Cahalan's 2019 book The Great Pretender has since cast serious doubt on how the study was run and reported. Most symptoms form a continuum. Plot how much sadness or worry people have and you get a bell curve with no natural gap; a diagnosis puts a line through it. Move the line and the number of cases moves with it. Interactive · Where do we draw the line? Cut-off, in standard deviations above average An invented, normally distributed "symptom load"; real conditions are not exactly bell-shaped and real diagnosis also asks about impairment and duration. This is a picture of an idea, not a screening tool. Without JavaScript: a cut-off 1.5 standard deviations above average labels about 6.7% of people; at 2 about 2.3%; at 1 about 15.9%. Whoever stands just below the line and whoever stands just above it are nearly alike. That is why a diagnosis is useful for getting help and for research, and why clinicians stress that it describes a pattern, not a separate kind of person. DSM-5-TR and ICD-11: two manuals, one field Two systems define what is counted. The DSM is published by the American Psychiatric Association; its fifth edition appeared in 2013 and a text revision, DSM-5-TR, in March 2022. The ICD is the World Health Organization's classification of all diseases; its eleventh revision, ICD-11, was endorsed by the World Health Assembly in May 2019 and came into effect on 1 January 2022, with a chapter on mental, behavioural and neurodevelopmental disorders. ICD is the official statistical standard for WHO member states, and many countries are still moving to ICD-11; DSM is used mainly in the United States and in much research around the world. They are mostly alike, but they differ at the edges, and the edges are revealing. Interactive · Same condition, two manuals Simplified summaries. Check the manuals for exact criteria. DSM-5-TR (APA, 2022)ICD-11 (WHO, in effect 2022) Who uses itMainly the US; widely in researchHealth systems and statistics worldwide StructureCategories with symptom checklistsClinical descriptions, with more dimensions CostSold as a bookFree online Do the manuals agree on who has what? Not as well as people assume. In the field trials for DSM-5, reported by Regier and colleagues in 2013, the agreement between clinicians interviewing the same patients on separate occasions was low for some common conditions, including major depression and generalized anxiety disorder, a result the APA itself acknowledged. Diagnoses are tools with real uncertainty, and critics have proposed more dimensional systems, such as the HiTOP model (Hierarchical Taxonomy of Psychopathology), which describes problems as positions on a few broad scales. The main conditions, in outline This is a map, not a guide to identifying anything. The two most common groups worldwide are anxiety and depressive disorders: the World Health Organization estimated in 2023 that about 1.2 billion people, roughly one in seven, were living with a mental disorder, including about 470 million with anxiety and 322 million with depression. Most conditions begin before the mid-twenties. GroupThe core of itMain treatments with evidence DepressivePersistent low mood or loss of interest, with sleep, energy and thinking changesPsychotherapy, medication, or both Anxiety, OCD, PTSDFear and worry out of proportion; obsessions and compulsions; reactions after traumaCBT with exposure; some medicines Bipolar and psychoticEpisodes of mania; losing contact with shared reality (psychosis)Medication is central, with psychosocial care NeurodevelopmentalAutism, ADHD, intellectual disability: differences present from early lifeSupport and adjustment, with medication for ADHD Eating and substance useDisordered eating; harmful use that is hard to stopSpecialist psychological, medical and community care PersonalityLong-standing patterns that cause trouble with self and othersSpecialist psychotherapy (for example, DBT) A diagnostic test is a screen of a different kind of question, and one piece of arithmetic helps whoever reads about any test. The chance that a positive result is a true case depends heavily on how common the condition is. This is Bayes' rule at work, the subject of Bayesian Thinking for Everyday Life (17/07). Interactive · What a positive result means How common the condition is Sensitivity: true cases the test catches Specificity: non-cases the test clears A made-up test for a made-up condition, applied to 1,000 people. It is about reasoning, not about any real instrument. Without JavaScript: with a 2% condition and a test that is 90% sensitive and 90% specific, about 16 of every 100 positive results are true cases; at 20% prevalence, about 69. This is why a short questionnaire is a first step, never a diagnosis: a positive result in a general population mostly leads to a conversation with a trained person, not a conclusion. Causes: the biopsychosocial picture "What causes depression?" has no single answer. The biopsychosocial model, introduced by George Engel in 1977, says that biology (genes, brain, illness), psychology (thinking habits, history) and the social world (poverty, relationships, culture) all matter, and they interact. The diathesis–stress model is the simplest version: a person has some vulnerability (the diathesis), and a disorder appears when stress pushes the total over a threshold. Two people with the same stress can have different outcomes. Twin studies suggest that genes matter, though not as destiny: heritability is high for schizophrenia (roughly 80%) and moderate for major depression (roughly 40%), and both leave plenty to the environment. Childhood adversity, loss, discrimination and poverty raise the risk of many conditions. And no single gene or chemical explains any common disorder. The "chemical imbalance" story deserves care. The serotonin theory of depression was popular for decades, and a 2022 umbrella review by Joanna Moncrieff and colleagues concluded that the research did not show a consistent link between serotonin and depression. Critics said the review was one-sided in its method and that it did not show that antidepressants do not work, and the researchers disagreed with them. What most clinicians now say is that medicines can help, that we do not understand exactly how, and that "depression is low serotonin" is a simplification. Interactive · Vulnerability plus stress Stress load Average vulnerability in the group A toy: vulnerability varies between people (standard deviation 15), and a disorder appears when vulnerability plus stress passes 100. The units are invented; the shape is the point. Without JavaScript: with average vulnerability 50 and stress 40, about 25% cross the threshold; at stress 60, about 75%. Look at the curve: stress is not a switch. It moves a share of the group over the line, and the people just below the line at one level of stress are the people over it at the next. That is both why prevention matters and why no person's illness is simply their own fault. Therapies and their evidence Psychotherapy has three broad families. Cognitive-behavioural therapy (CBT), developed from the 1960s by Aaron Beck and others, works on the link between thoughts, feelings and behaviour, and on facing feared things in steps; it has the largest research base. Psychodynamic therapies, descended from Freud, look at patterns that began early in life; they have evidence for several conditions, though less of it and more argument over its quality. Humanistic approaches, from Carl Rogers, put the relationship at the centre. Newer methods include dialectical behaviour therapy for severe emotional difficulty, and EMDR for trauma, which is recommended by guidelines but debated over what part of it does the work. How do you say a therapy "works"? You compare outcomes to a control group and report the gap in standard deviations, the effect size (Cohen's d). By convention 0.2 is small, 0.5 medium and 0.8 large. The picture below turns that into something you can feel. Interactive · What an effect size looks like Small 0.2Medium 0.5Large 0.8 Effect size, d Two normal curves with equal spread. "Chance a treated person does better" is the probability of superiority, Φ(d ÷ √2). Without JavaScript: d = 0.2 means a randomly chosen treated person does better than a random control about 56% of the time; d = 0.5, about 64%; d = 0.8, about 71%. Notice that even a "large" effect leaves wide overlap: many people in the control group do better than the average treated person. That is true of nearly everything in this field and it is not a reason for despair. A Lancet network meta-analysis in 2018 (Andrea Cipriani and colleagues; 522 trials, 116,477 people) found that all 21 antidepressants it studied beat placebo in short-term trials, with odds ratios between about 1.4 and 2.1, so the benefit is real but modest and not the same for everyone. Psychotherapy and medication have each been shown to help in depression, and for many anxiety conditions psychotherapy does as well or better, with longer-lasting benefit once treatment stops. A caution. Published results flatter treatments. In a 2008 analysis in the New England Journal of Medicine, Erick Turner and colleagues compared FDA records with journal articles for antidepressant trials, and found the journal literature looked about 94% positive while the FDA files showed about half. Pre-registration and publishing negative results have helped, but trial size, who is excluded and who drops out still affect how well a result transfers to a real patient. Another persistent finding is that the relationship between patient and therapist predicts outcomes across many methods, which is part of why a good fit matters. Whether all therapies are equally good (the "Dodo bird verdict") is still argued. Global mental health: who gets care Most of the world's research comes from rich countries, and most of the world's people with mental disorders live elsewhere. The World Health Organization reports that only about a third of people with depression receive formal care, and the gap is far wider in poor countries. Its Mental Health Atlas 2020 put the global median at 13 mental health workers per 100,000 people, with fewer than two per 100,000 in low-income countries and over 60 in high-income countries. Interactive · One worker, how many people? Low-incomeGlobal medianHigh-income Mental health workers per 100,000 people The one-in-seven figure is WHO's 2023 estimate; the worker figures are from the 2020 Atlas (the low and high ends as reported in secondary summaries). Not everyone with a disorder needs a specialist, so this is a measure of scale, not of waiting lists. The bar is drawn on a square-root scale. Without JavaScript: at 13 workers per 100,000, about 1,100 people with a disorder (one in seven) per worker; at 60, about 240; at 1.4, over 10,000. One response is to take treatment out of the clinic. The Friendship Bench, created in Harare, Zimbabwe, by the psychiatrist Dixon Chibanda, trains lay health workers, often older women known as "grandmothers", to give six sessions of problem-solving therapy on a bench in the grounds of a clinic. In a 2016 randomised trial in JAMA with 573 adults, six months later about 14% of the Friendship Bench group still had symptoms of depression on the measure used, against about 50% of those who got usual care. It spread to several other countries. The WHO's mhGAP programme and Vikram Patel's work on task-sharing follow the same logic: train more people to deliver a limited, evidence-based set of skills. Culture shapes how distress shows. The DSM now lists "cultural concepts of distress": ataque de nervios, common in Latin America and the Caribbean; kufungisisa ("thinking too much") among the Shona of Zimbabwe, which can describe worry or depression; susto, fright that is believed to cause illness. These do not map neatly onto Western categories, and a good clinician asks about meaning before sorting a person into a diagnosis. What a graduate just sees Most symptoms form a continuum; a diagnosis puts a line through it, so people on either side of it are alike. The manuals disagree at the edges; a label is a tool with error, not a discovery. A positive on any test, for a rare condition, is mostly a false alarm until someone checks. Causes are many and interacting; one cause such as "a chemical" or "a gene" is a simplification. A treatment's effect is an average with wide overlap, so a good result for most is not a promise for you. Published trials look better than the full record; ask what was left out. Most people who need care get none, and trained lay workers can do a lot of good. Distress has local language; ask what it means to the person. Where it shows up next Organizational Behavior (34/53): burnout, workplace mental health and the evidence behind employee wellbeing programmes. Developmental Psychology (35/24): how early adversity and attachment shape later risk. Biological Psychology (35/20) and Personality & Individual Differences (35/26): the brain and the traits that clinical work builds on. Public Economics (34/49): who pays for care, and why prevention is cheaper than it looks. In the world: school counselling, crisis lines, workplace insurance, and the headlines that say a drug or an app "cures" something. If you want to go deeper 📘 Textbook. David Barlow and Mark Durand, Abnormal Psychology: An Integrative Approach (Cengage, many editions), or Ronald Comer, Abnormal Psychology (Worth). Both are common undergraduate texts. 📘 Reference. American Psychiatric Association, DSM-5-TR (2022), and the free online ICD-11 browser at icd.who.int. 📘 Gentler. Allen Frances, Saving Normal (2013), a critical view of diagnostic inflation by the man who chaired the DSM-IV task force, and Susannah Cahalan, The Great Pretender (2019). 🌐 Free course. Open Yale Courses PSYC 110, Introduction to Psychology (Paul Bloom) includes lectures on mental illness and its treatment. The free Noba textbooks have clinical modules. The takeaway Disorders are patterns on a continuum, defined by distress and impairment, and described by two manuals that agree on much and not on all. Causes are biological, psychological and social at once. Several therapies help on average, with real uncertainty, and the largest problem worldwide is not choosing a treatment but getting any. Try this: the next time a headline claims a new therapy or drug works, find three things: the size of the effect, the comparison group, and whether the result was reproduced. Sources & Further Reading Sources. American Psychiatric Association, DSM-5 (18 May 2013) and DSM-5-TR (18 March 2022; includes prolonged grief disorder); WHO, ICD-11 (endorsed by the World Health Assembly on 28 May 2019, in effect 1 January 2022; complex PTSD 6B41, gaming disorder 6C51; gender incongruence in the sexual-health chapter; personality disorder graded by severity). Wakefield, American Psychologist 1992; Rosenhan, Science 1973, and Cahalan 2019; homosexuality removed from the DSM in 1973 and from the ICD in 1990. Regier et al., "DSM-5 Field Trials in the United States and Canada, Part II," American Journal of Psychiatry 2013. WHO fact sheet on mental disorders (2023 figures: about 1.2 billion people, 470 million with anxiety, 322 million with depression; about a third of people with depression receive formal care): who.int/news-room/fact-sheets/detail/mental-disorders. WHO, Mental Health Atlas 2020 (global median 13 workers per 100,000; WHO news release of 8 October 2021); low and high ends as given in secondary summaries and not checked against the report. Engel, Science 1977; Moncrieff et al., Molecular Psychiatry 2022, and the critical comment by Jauhar and colleagues (2023); Cipriani et al., The Lancet 2018 (522 trials, 116,477 participants, odds ratios 1.37 to 2.13); Turner et al., New England Journal of Medicine 2008. Chibanda et al., "Effect of a primary care-based psychological intervention on symptoms of common mental disorders in Zimbabwe," JAMA 2016 (573 participants, 24 clinics; depression symptoms 14% against 50% at six months). Twin heritability of about 80% for schizophrenia and about 40% for major depression: Sullivan and colleagues (2000, 2003), given as round figures. Computed in node from the code on this page: prevalence above a cut-off of 1.5 SD is 6.7%; PPV 15.5% at 2% prevalence and 69.2% at 20%; probability of superiority 55.6%, 63.8% and 71.4% at d = 0.2, 0.5 and 0.8. 📘 Barlow and Durand, Abnormal Psychology 🌐 findahelpline.com 🌐 Open Yale Courses PSYC 110 Step 11. Behavioral Economics: How People Actually Choose --------------------------------------------------------- Why: Biases and decisions, in money. Standard economics assumes calm, consistent, self-interested calculators. Behavioral economics measures how real people depart from that, and, just as important, which of its famous findings have survived being checked. You sign up for a gym at 8 a.m. in January, full of resolve, and pay monthly for a year. You go four times. A shop shows a jacket "was £200, now £120" and it suddenly feels like a bargain, though you had no idea what it was worth a minute ago. A company enrolls new staff in a pension plan automatically, and far more people end up saving, although nothing about their finances changed. None of this is irrational in the sense of being random. It is irrational in a predictable way, and predictable is what an economist can use. Behavioral economics collects those regular departures from the textbook chooser and builds them into models. It is also a field that has been through a hard decade, in which many headline results were re-tested and some did not survive. This piece teaches both: the ideas, and the report card. The course in one box Who takes itEconomics majors, usually in year 3 after microeconomics and statistics; marketing, psychology and business-school programmes teach it as consumer behaviour or decision-making. MIT, for example, offers a course called Psychology and Economics (14.13). What it is forPredicting where real choices depart from the rational-agent benchmark, and designing prices, defaults and policies that work for real people. Comes afterMicroeconomics (34/42) and Probability & Statistics (34/05). Leads toConsumer Behavior (34/58), and it feeds Pricing (34/61) and Marketing Management (34/57). Econometrics (34/44) is the toolkit for judging which of its findings are real. What this piece gives youYou will not be able to design a behavioural experiment. You will know the main effects, roughly how big they are, and which classics were overturned. Easier first reads: The Biases That Fool Everyone (17/02), How to Make a Hard Decision (17/05) and Why You Understand Less Than You Think (17/10). This piece goes further by asking how much each bias is worth and whether it held up. Bounded rationality: shortcuts that usually work The textbook consumer weighs every option, knows her own tastes and never makes arithmetic mistakes. Herbert Simon argued in the 1950s that this ignores the obvious: minds are limited, time is short, information is incomplete. People do not optimise, they satisfice: search until something is good enough, then stop. He called this bounded rationality. Daniel Kahneman and Amos Tversky later catalogued the rules of thumb, or heuristics, that people use, and the systematic errors each one produces. Judging how likely something is by how easily examples come to mind (availability), by how much it resembles a stereotype (representativeness), or by starting from a number you were handed (anchoring), all work most of the time, and all fail in patterned ways. Kahneman received the 2002 economics Nobel prize for this work. A worked example, from the "cognitive reflection" literature: a bat and a ball cost £1.10 together; the bat costs £1.00 more than the ball. How much is the ball? The quick answer is 10p. Check it: ball 10p, bat £1.10, total £1.20. If the ball costs x, the bat costs x + 1.00, so 2x + 1.00 = 1.10 and x = 5p. The wrong answer comes fast and feels right; the right one takes a second look. Biases are mostly this: a fast process that is usually fine, with no alarm bell when it is not. The rest of the piece is a tour of the best-studied ones, with a growing sense of how much to trust each. Prospect theory: gains, losses and the kink Kahneman and Tversky's 1979 paper in Econometrica proposed prospect theory as a description of how people value risky outcomes. Three ingredients. First, people judge outcomes as gains or losses from a reference point (usually where they are now), not as totals of wealth. Second, diminishing sensitivity: the difference between winning 0 and 100 feels larger than between winning 1,000 and 1,100. Third, loss aversion: a loss hurts more than the same gain pleases. Their 1992 estimates were roughly α ≈ 0.88 for curvature and λ ≈ 2.25 for loss aversion. v(x) = xα for gains, v(x) = −λ·(−x)α for lossesλ above 1 makes the curve steeper for losses; α below 1 flattens it as amounts grow Interactive · Would you take the coin flip? Loss aversion λ Curvature α Win if heads (lose 100 if tails) A simplified prospect-theory value function, ignoring probability weighting. The values are in arbitrary "felt" units. At λ = 1 and α = 1 the chooser is the textbook risk-neutral one. Without JavaScript: with λ = 2.25 and α = 0.88, a 50:50 chance to lose 100 is only worth taking if the win is at least about 251; with λ = 2 and α = 1 you need exactly 200. At the default settings a coin flip that risks 100 and offers 150 has a positive average (+25 in money) but a negative felt value. Slide the win up to about 250 and it flips. People often refuse such bets, and prospect theory says why. How robust is loss aversion itself? That is an unfinished argument, taken up at the end. Anchoring and framing: the same facts, different pull In the best-known anchoring experiment, Tversky and Kahneman (1974) spun a wheel that stopped on 10 or 65, then asked people what percentage of United Nations members are African countries. People who saw 10 gave a typical answer near 25%; those who saw 65 answered near 45%. A random number pulled the estimate. The sliding tool below interpolates between those two reported medians (an illustration of the size of the pull, not data beyond them). Interactive · The anchor's pull Number the wheel showed Line through the two reported median estimates (anchor 10 gives 25; anchor 65 gives 45). Between them, every 1 point of anchor moves the typical estimate by about 0.36 points. The anchoring index is the shift in estimates divided by the shift in anchors: (45 − 25) ÷ (65 − 10) ≈ 0.36. Anchoring is one of the sturdier findings in the field, though a few studies with obviously random anchors have not found it. Framing is the cousin: Tversky and Kahneman's 1981 "Asian disease" problem describes a programme for 600 people at risk either as "200 will be saved" or as "400 will die", and both describe the same outcome. In the original, a majority took the sure option when it was framed as lives saved and a majority took the gamble when framed as deaths. Framing effects have been re-tested widely, and the direction usually holds; the size depends heavily on the wording and the people asked. Present bias: today versus later Standard theory discounts the future at a steady rate, so your preference between "£100 on day d" and "£110 on day d + 7" stays the same as the days pass. Real people do not. Richard Thaler found in 1981 that people would be indifferent between $15 now and $30 in three months, $60 in a year or $100 in three years, implying discount rates that fall steeply with the delay. George Ainslie found many people preferred $50 now to $100 in six months but $100 in nine months to $50 in three: the same choice, six months further away, reverses. David Laibson (1997) captured this in the quasi-hyperbolic model: everything after today gets an extra one-off discount β below 1. value of reward x at delay t = β · δt · x for t > 0, and x if nowδ is ordinary patience per day; β below 1 is "present bias" Interactive · Plan today, choose tomorrow Present bias β (1 = none) Days until the smaller reward Choice: £100 on that day, or £110 a week later. Patience δ = 0.995 per day (about 3.4% lost per week). Invented parameters for illustration. With β = 0.7 your plan today is to wait for the £110 whenever the smaller prize is a day or more away. When the day arrives and £100 is available now, you take it, against your own plan. That plan-then-break pattern is the gym membership, the unfinished diet, the unopened savings account. People who know they are like this buy commitment devices: a locked savings account, a bet with a friend, deleting the app. Set β = 1 and the reversal vanishes. Defaults and nudges: design the easy path If people are inattentive and slightly present-biased, whatever happens when they do nothing matters enormously. Madrian and Shea (2001) studied a large US company that switched its 401(k) pension plan to automatic enrolment, with the plan's economic terms unchanged: participation was significantly higher under automatic enrolment, and many employees stayed at the default contribution rate and fund. National organ-donation rates tell a similar story: about 12% consent in opt-in Germany against 99.98% in opt-out Austria is the oft-cited contrast, though it mixes in other causes, and Spain, with the highest donor rate, credits its transplant coordination more than its law. Interactive · Opt-in or opt-out? Share of people who actively decide Share who truly want to join A toy model: active deciders follow their own wish; everyone else stays at the default. Real take-up depends on many more things. The toy model shows the logic: the less attentive the population, the more the default decides, and the fewer people are choosing at all. This is the idea behind the nudge, a term Richard Thaler and Cass Sunstein made famous in their 2008 book: change the "choice architecture" without forbidding anything. It is also the cue for a warning. Whether nudges work at scale is disputed (report card below), and the ethics of steering people are argued too: critics worry about manipulation, supporters reply that every choice has some default, so someone has to set it. Social preferences: the ultimatum game Standard theory predicts people care only about their own payoff. In the ultimatum game (Güth, Schmittberger and Schwarze, 1982), a proposer splits 100 with a responder who can accept or reject; on rejection both get nothing. A pure money-maximiser would accept 1, so the proposer would offer 1. Real responders do not. Across 37 studies, proposers typically offered about 40%, and offers below about 30% were often rejected, so responders pay to punish what they see as unfair. Fifteen small-scale societies studied by Henrich and colleagues showed that what counts as fair differs across cultures, so the "typical" figures belong to the places where most experiments have been run. Interactive · Make the offer Your offer (out of 100) Typical minimum responders accept How much responders differ Each responder has a private minimum, spread as a bell curve around the typical value. Invented parameters: the point is the shape of the trade-off, not a forecast. Set the minimum and the spread to near zero (a textbook responder): the best offer collapses to about 3. Put them near what the studies describe and the best offer jumps to about 39, because a stingy offer risks everything. A proposer who offers fairly may be motivated by fairness or by fear of rejection, and the game alone cannot say which. What held up, what shrank, what failed Behavioral economics grew out of psychology, and psychology went through a replication crisis. In 2015 the Open Science Collaboration re-ran 100 published psychology studies and found about 36% replicated successfully by their main criterion. Economics did better but not perfectly: a 2016 project re-ran 18 laboratory experiments from top economics journals, and 11 (about 61%) replicated; a 2018 project re-ran 21 social-science papers from Nature and Science, and 13 (about 62%) replicated, typically with smaller effects. The report card below sorts the findings used in this piece and a few famous neighbours. Use the buttons to filter it. Interactive · The report card All Held up Shrank Failed Disputed A few lessons. Directions hold more often than sizes: defaults, anchors and present bias point the way the theory says, but their effects in the field are often smaller than the first lab study. Big, flashy, one-off findings are the riskiest: ego depletion and social priming were built on small samples and clever stories. Not every "bias" is one idea: loss aversion, the endowment effect and the status quo bias may be one force or three, and experts differ. And published averages exaggerate, for the reason shown in Econometrics (34/44): significant results survive the journal filter, and lucky overestimates are the ones that pass it. What a graduate just sees People use shortcuts that usually work and sometimes fail in patterned ways; the pattern is what makes it a science. Outcomes are judged from a reference point, and losses weigh more than equal gains, probably about twice, but the figure depends on context. A number you are handed pulls your estimate toward it, even when it is irrelevant. The same facts framed as gains or losses can flip a choice. Present bias makes plans and actions disagree; commitment devices exist because people know it. The default is a choice someone made for you; when people are inattentive, it decides. People will pay to punish unfairness, and fairness norms differ across societies. Ask of any famous finding: replicated? how big? in whom? Direction survives more often than size. Where it shows up next Consumer Behavior (34/58): reference prices, anchoring on list prices and "was/now" framing are everyday marketing. Pricing (34/61): decoys, round numbers and subscription defaults rely on these effects, and the replication record tells you which to trust. Public Economics (34/49): automatic enrolment and opt-out donation are policy tools, with the nudge debate attached. Strategy & Competition (34/54) and Game Theory (34/45): the ultimatum game is a game theory result that real people do not follow. In the world: pension design, fundraising, health messaging, app design and every "limited time offer". If you want to go deeper 📘 Textbook. Nick Wilkinson and Matthias Klaes, An Introduction to Behavioral Economics (Palgrave). A standard undergraduate text. 📘 Reference. Colin Camerer, George Loewenstein and Matthew Rabin (eds.), Advances in Behavioral Economics (Princeton, 2004), for the research papers behind the field. 📘 Gentler. Richard Thaler, Misbehaving: The Making of Behavioral Economics (Norton, 2015), and Daniel Kahneman, Thinking, Fast and Slow (2011). Read both with the report card in mind, as some of the studies in the second have since been questioned. 🌐 Free course. MIT OpenCourseWare 14.13 Psychology and Economics (Spring 2020, Prof. Frank Schilbach). The takeaway Real choosers rely on shortcuts, judge from reference points, discount the immediate over the later, follow defaults and care about fairness. Those departures are real and often predictable, but sizes are smaller and less universal than the first headlines claimed, and a few famous results did not survive. Trust the direction, doubt the magnitude, and check the replication. Try this: this week, find one default in your life (a subscription, a notification setting, a pension contribution) and decide deliberately whether it is the one you would have chosen. Sources & Further Reading Sources. Simon on bounded rationality and satisficing (1950s); Tversky & Kahneman, "Judgment under Uncertainty: Heuristics and Biases," Science 1974 (wheel-of-fortune anchoring: anchors 10 and 65, typical answers about 25% and 45%); Kahneman & Tversky, "Prospect Theory: An Analysis of Decision under Risk," Econometrica 47(2), 1979; Kahneman won the 2002 economics Nobel. Value-function parameters α ≈ 0.88 and λ ≈ 2.25 are the Tversky & Kahneman (1992) estimates; they vary by sample and are used here only as an illustration. Tversky & Kahneman, "The Framing of Decisions and the Psychology of Choice," Science 1981. Thaler, "Some Empirical Evidence on Dynamic Inconsistency," 1981 ($15 now against $30 in 3 months, $60 in 1 year, $100 in 3 years); Ainslie's $50/$100 reversals; Laibson, "Golden Eggs and Hyperbolic Discounting," 1997. Madrian & Shea, NBER Working Paper 7682 (2000), published 2001: participation "significantly higher under automatic enrollment." Organ-donation contrast and the caveat about other causes: Wikipedia, "Organ donation." Thaler & Sunstein, Nudge (2008). Nudge effectiveness: Mertens et al. (PNAS 2022) found moderate effects; Maier et al. (PNAS 2022) found no evidence after adjusting for publication bias; DellaVigna & Linos (Econometrica 2022) found smaller effects in nudge-unit trials than in published papers. Güth, Schmittberger & Schwarze, 1982; Oosterbeek, Sloof & van de Kuilen 2004 (37 studies: offers about 40%, offers below about 30% often rejected); Henrich et al. on small-scale societies. Replication: Open Science Collaboration 2015 (36%); Camerer et al. 2016 (11 of 18) and 2018 (13 of 21); ego depletion: Hagger et al. 2016 (2,141 participants, no evidence) and Vohs et al. 2021 (d = 0.06); marshmallow test: Mischel & Ebbesen 1970 (32 children), Watts, Duncan & Quan 2018; loss aversion: Gal & Rucker 2018, Mrkva et al. 2020, Yechiam & Zeif 2025; choice overload: Iyengar & Lepper 2000, Scheibehenne et al. 2010, Chernev et al. 2015. Computed in node from the code on this page: break-even win in the coin flip = λ1/α × 100 = 251 (200 at λ = 2, α = 1); anchoring index 20 ÷ 55 = 0.36; bat and ball 5p; the toy ultimatum game's best offers (39 and 3). 🌐 MIT OCW 14.13 Psychology and Economics 📘 Thaler, Misbehaving 📘 Kahneman, Thinking, Fast and Slow Step 12. Organizational Behavior & Leadership: People Are the System -------------------------------------------------------------------- Why: Psychology at work. Why some teams speak up and others go quiet, what actually moves people to work, and which famous findings about people at work did not survive being repeated. Two teams get the same brief, the same budget and equally talented people. One ships, learns and enjoys it. The other argues in the corridor, nods in meetings and quietly fails. Nothing in the spreadsheet explains the difference. Anyone who has worked in both knows it is real. Organizational behavior is the study of that difference: what moves people to work, how groups think, why some teams speak up and others don't, who gets listened to, and what it takes to lead. It is also a field that has had a rough decade. Several of its most famous findings did not survive being re-run. This piece teaches the parts that held up, and says plainly which famous ones did not. The course in one box Who takes itBusiness students, usually in year 2, under names like Organizational Behavior, Management or Leadership. Management is part of Wharton's Business Fundamentals core, and leadership is part of the first-year Required Curriculum at Harvard Business School (see 00_MAJORS). Psychology and engineering students meet a lighter version. What it is forUnderstanding why people in organisations do what they do, so you can design work, run meetings and lead teams without relying on myths. Comes afterNothing in particular; it builds on ordinary life in groups. Leads toStrategy & Competition (34/54) and Entrepreneurship (34/56) in the Business Administration major; Marketing Management (34/57) lists it too, because marketing is teams and persuasion inside the firm. It links to Game Theory (34/45) for negotiation, Behavioral Economics (34/46) for biases, and Software Engineering (34/31) and Systems Engineering (34/22), where teams are the real system. What this piece gives youYou will not become a coach or a CEO. You will know which claims about people at work are solid, which are folklore, and how to play a negotiation. Easier first reads: The Biases That Fool Everyone (17/02), Thinking in Systems (17/06) and Negotiation for Normal People (16/10). Motivation: what holds up and what doesn't Why do people work hard? The two theories with the most evidence behind them are less glamorous than the ones on posters. Goal-setting theory comes from Edwin Locke and Gary Latham. Across decades of laboratory and field studies, people given a specific, difficult goal do better than people told to "do your best", provided they accept the goal, have the ability and get feedback. A 1981 review reported that about 90% of studies found the effect. It has limits: on a complex task you have never done, a specific performance target can backfire and a learning goal ("find three strategies") works better, and Ordonez and colleagues argued in 2009 that goals can narrow attention and invite cutting corners. Locke and Latham answered in print, and the debate continues. Self-determination theory, from Edward Deci and Richard Ryan, says that people are most motivated and well when three needs are met: autonomy (acting from choice), competence (getting better) and relatedness (belonging). Its best-known result is that paying someone for something they already enjoy can reduce their interest afterwards: Deci found it for money in 1971, while praise did not hurt. How large and general that effect is remains argued, but the practical lesson, that control crushes interest and choice sustains it, is widely used. What does not hold up is a surprisingly long list. Abraham Maslow's pyramid of needs is the clearest case: Maslow never drew a pyramid. It is credited to a 1960 article by Charles McDermid, and while later work supports universal human needs, the strict stair-step order is not supported. Power posing and ego depletion are the other famous casualties. Test yourself: for each classic claim, guess how it fared. Interactive · Did it replicate? Holds up Holds, with limits Did not hold Mixed or misattributed Next claim Verdicts follow the sources listed at the end. "Holds up" means supported by several independent studies or reviews, not that it is settled for ever. The habit this teaches is the main thing the course gives: when someone says "research shows", ask how big the study was, whether it was repeated by someone else, and who stands to gain from the claim. Decisions in groups: groupthink, and the wisdom and madness of crowds Groups feel wiser than individuals, and sometimes are. In 1907 Francis Galton collected about 800 guesses at a country fair in Plymouth for the weight of a slaughtered ox. The median was 1,207 pounds. The true weight was 1,198: within 1%. James Surowiecki's 2004 book The Wisdom of Crowds made the idea famous and named the conditions: diverse views, guesses made independently, and a question with a checkable answer. Averaging cancels errors that are random; it cannot cancel errors that everyone shares. The opposite failure is Irving Janis's groupthink (1972, revised 1982): tight, insulated groups under stress that seek harmony over truth. He applied it to the Pearl Harbor warnings, the Bay of Pigs invasion and the escalation of the Vietnam War, and listed eight symptoms such as an illusion of invulnerability and pressure on dissenters. His remedies are still good practice: a designated devil's advocate, a leader who speaks last, outside experts, and independent sub-groups. As a scientific theory it has fared less well. Lab and field evidence is mixed, and later researchers (Kramer in 1998, Baron in 2005) questioned how Janis read his case studies. Treat it as a checklist of risks, not a law. The crowd can also go mad when people see each other. Jan Lorenz and colleagues showed in a 2011 PNAS paper that letting people see others' estimates narrowed the spread of answers without making the average more accurate. You can reproduce the logic below: each simulated person starts with their own noisy guess of the ox's weight, then some of them move toward the running average of earlier guessers. Interactive · Guess the ox Crowd size How much people lean on earlier answers New crowd Illustrative model: each person's private guess is the true weight (1,198 lb) times a random factor with about 25% typical error, no shared bias. With influence w, a person's answer is (1 − w) × own guess + w × the average of all earlier answers. Not data from a real study. Turn the influence up: the dots huddle together and the crowd looks sure, yet its average gets worse, because early guessers set the tone for everyone. Agreement is not evidence. That is why good decision processes collect opinions before discussion. Teams: psychological safety The most useful single idea in modern team research is Amy Edmondson's psychological safety: the shared belief that you can take an interpersonal risk, such as asking a question, admitting a mistake or disagreeing with the boss, without being punished or humiliated. She defined it in a 1999 study of work teams (Administrative Science Quarterly 44(2)) after a puzzle in hospitals: in an earlier study, units with better leadership appeared to make more errors. The explanation was that people on those units felt able to report them. Teams that report nothing are not safer, only quieter. Google's internal Project Aristotle, begun around 2012, went looking for what made its own teams effective, expecting to find the right mix of skills and personalities. What it reported instead was that who was on the team mattered less than how members treated each other. Its write-up named five dynamics, with psychological safety the most important: then dependability, structure and clarity, meaning, and impact. Be careful about what that proves. It is one company's internal analysis, not a controlled experiment, and the safety link is correlational. Independent academic reviews do support a positive link between psychological safety and learning and performance, but "safe" does not mean "soft": the strongest teams also hold high standards. P(hear it) = 1 − (1 − p)kk people notice a problem, each speaks up with chance p Here is the simple arithmetic of a quiet team. Suppose one junior person spots a problem and would speak up only 30% of the time. The team hears about it 30% of the time: seven times in ten, the problem stays hidden. If three people notice it, each at 30%, the chance rises to about 66%. Safety is not niceness; it is the difference between a team that hears its own bad news and one that hears it from customers. Interactive · Does the team hear about the problem? Chance each person speaks up People who noticed Meetings before it is too late A toy model: independent people, same chance every meeting. Real speaking-up is not independent (one person's silence teaches the next). Illustrative. Power and influence People often follow because they must, can be rewarded, admire someone or simply believe the person knows. In 1959 the social psychologists John French and Bertram Raven sorted this into five bases of power: legitimate (the role gives the right to ask), reward, coercive (the power to punish), expert (knowing more) and referent (being liked or admired). Raven later added a sixth, informational power. The list is a taxonomy, not a law, but it is a good diagnostic: the base decides how long the influence lasts and what it costs. Coercion works only while you are watching; expertise and admiration work when you are not. Interactive · Name the base of power Legitimate Reward Coercive Expert Referent Next scenario Position in a network matters as much as rank: the person who connects two groups, or who everyone asks for help, has influence no title gives. And one popular shortcut did not survive: the idea that standing in a "power pose" changes your hormones and your behaviour. Co-author Dana Carney said in 2016 she no longer believed the effects were real. Looking confident may help how you feel; it does not rewrite your biology. Culture: what gets rewarded is what you get Edgar Schein, whose 1985 book on organizational culture is a standard text, described culture as three layers. On top are artifacts: dress, office layout, rituals, slogans. Beneath are espoused values: what the company says it believes. At the bottom are basic assumptions: the unspoken rules people actually follow, which you learn by watching what happens when values and targets collide. When the stated value is "safety first" and the bonus is paid on output, the assumption is output first. Steven Kerr's 1975 article "On the Folly of Rewarding A, While Hoping for B" is the classic version: organisations announce teamwork and pay for individual results, announce quality and pay for volume, then wonder why people do what is paid for. The practical test of a culture is the answer to a few questions. Who got promoted last year? What was the last bad news someone delivered, and what happened to them? What would a new hire learn from the first week about what is tolerated? Culture is not a poster; it is what the reward system, the leaders' behaviour and the stories people tell all make rational. Leadership: what the evidence says The oldest leadership experiment is Kurt Lewin, Ronald Lippitt and Ralph White's 1939 study of children's clubs led in autocratic, democratic and laissez-faire styles. It popularised the three styles and the finding that democratic groups were generally more satisfied. It was small, and adults were acting out styles; it is a landmark, not a proof of a best style. Later research moved on. Trait searches (are leaders taller, more extroverted, more intelligent?) found modest links at best. Contingency theories, such as Fred Fiedler's of the 1960s, argued that the best style depends on the situation. The model with the most research behind it is transformational leadership, from James MacGregor Burns and Bernard Bass: leaders who inspire a shared vision, consider individuals and challenge assumptions. Meta-analyses such as Judge and Piccolo's (2004) find it linked to followers' satisfaction and to leaders' rated effectiveness, though contingent reward, a "transactional" practice of clear rewards for results, also shows positive links. Critics (including Daan van Knippenberg and Sim Sitkin, 2013) point to a problem in the measures, which overlap with the outcomes they predict, and to the difficulty of showing that the style causes the result rather than following success. What a reader can safely carry away: leaders matter, but less than heroic stories suggest; clear direction, fairness, support and honest feedback have the most consistent backing; and people credit leaders for results they only partly controlled, a bias James Meindl and colleagues called the "romance of leadership" in 1985. Negotiation basics: know your BATNA In 1981 Roger Fisher and William Ury, of the Harvard Negotiation Project, published Getting to Yes. Its central tools are three. Separate people from the problem. Focus on interests (why you want something) rather than positions (what you say you want). And know your BATNA: your Best Alternative To a Negotiated Agreement, the best thing you can do if the talks fail. The idea echoes the "disagreement point" in John Nash's bargaining theory. Your BATNA is your real power, and also your walk-away line: never accept less than it. Your reservation price is the worst deal you would take; the other side has theirs. If your reservation price is below theirs, a zone of possible agreement (ZOPA) exists between the two, and the negotiation is about how to divide it. If not, no deal is available and the right move is to walk. Research on anchoring (Galinsky and Mussweiler, 2001) finds that whoever makes the first offer often pulls the final price their way, which is why preparation matters. Interactive · Sell the car Dealer offers $3,500 Dealer offers $4,500 Dealer offers $5,500 Your asking price Ask this priceAccept buyer's bidTake the dealer's offerNew buyer A teaching model. The buyer has a hidden limit between $4,000 and $7,000. Each round the buyer closes 30% of the gap to that limit, and accepts any ask up to halfway between its bid and its limit. Real buyers are less predictable. Illustrative. Play it a few times. The best asks sit just under the buyer's limit but not above it, which you can only judge from how the bids move, and sometimes the buyer's limit is below your dealer's price, so the winning move is to leave. Negotiation is mostly preparation: knowing your alternative before the conversation starts. What a graduate just sees "Research shows" is a prompt to ask: how large, repeated by whom, and what survived? Specific goals help, but narrow goals can corrupt what they don't measure. Control kills interest; choice, competence and belonging sustain it. A group that agrees fast is not necessarily right; collect opinions before the discussion. Silence is not safety. A team that reports no problems may be one that cannot. Ask which base of power a person uses: it tells you how long their influence will last. Culture is what gets rewarded and tolerated, not what is on the wall. Know your BATNA before you sit down; it is your walk-away line and your leverage. Where it shows up next Strategy & Competition (34/54) and Entrepreneurship (34/56): a strategy has to be carried out by people, and a start-up is mostly a team surviving uncertainty. Behavioral Economics (34/46) and Game Theory (34/45): the biases and the bargaining model under the people-side lessons here. Software Engineering (34/31) and Systems Engineering (34/22): code reviews, post-mortems and design reviews work or fail on psychological safety. Marketing Management (34/57): persuasion inside the firm, agency relationships and the buyer-seller negotiation. In the world: hiring and promotion, running meetings, performance reviews, mergers, hospitals and airline crews, and every salary talk you will ever have. If you want to go deeper 📘 Textbook. Stephen Robbins and Timothy Judge, Organizational Behavior (Pearson). The most widely assigned undergraduate text, with chapters on every topic here. 📘 Textbook. The free OpenStax Organizational Behavior text, written for introductory courses and available online at no cost. 📘 Gentler. Amy Edmondson, The Fearless Organization (Wiley, 2018) on psychological safety; Roger Fisher and William Ury, Getting to Yes (1981) on negotiation. 🌐 Free course. MIT OpenCourseWare 15.311 Organizational Processes (Fall 2003): a graduate Sloan course in organisations and leadership. The takeaway People at work are moved by choice, competence and belonging more than by pyramids and poses. Groups are wise when members think independently and foolish when they copy each other. Teams learn when it is safe to speak. Culture is what gets rewarded. Lead with clarity and fairness, and walk into any negotiation knowing your alternative. Trust the findings that have been repeated, and be wary of the ones that made a good story. Try this: at your next meeting, write down your view before anyone speaks, and ask two others to do the same. Compare. Notice how much of the later discussion simply moved toward the first confident voice. Sources & Further Reading Sources. Goal setting: Locke & Latham, "Building a practically useful theory of goal setting and task motivation", American Psychologist (2002); Ordonez et al., "Goals Gone Wild" (2009). Self-determination theory: Deci (1971); Deci & Ryan (1985). Maslow's pyramid: attributed to McDermid, Business Horizons (1960); Tay & Diener (2011). Power posing: Carney, Cuddy & Yap (2010); Ranehill et al. (2015); Carney's 2016 statement. Ego depletion: Hagger et al. (2016); Vohs et al. (2021). Groupthink: Janis (1972, 1982); Kramer (1998); Baron (2005). Crowds: Galton (1907); Surowiecki (2004); Lorenz et al., PNAS (2011). Psychological safety: Edmondson, ASQ 44(2), 1999; Google re:Work on Project Aristotle. Bases of power: French & Raven (1959). Culture: Schein (1985); Kerr, Academy of Management Journal (1975). Leadership: Lewin, Lippitt & White (1939); Burns (1978); Bass (1985); Judge & Piccolo (2004); Meindl, Ehrlich & Dukerich (1985). Negotiation: Fisher & Ury (1981); Galinsky & Mussweiler (2001). Replication facts were checked against the Wikipedia summaries of these studies, which cite the papers. Interactive numbers were computed and checked in code; the data are illustrative. 🌐 MIT OCW 15.311 Organizational Processes 📘 Robbins & Judge, Organizational Behavior 📘 Edmondson, The Fearless Organization 📘 Fisher & Ury, Getting to Yes Step 13. Consumer Behavior: How People Notice, Want, Choose and Regret ---------------------------------------------------------------------- Why: Psychology in the market. Nobody weighs every option. People notice a sliver of what is in front of them, follow each other more than they admit, and regret in predictable ways. A graduate also knows which famous "findings" have held up and which were never real. You walk into a supermarket for milk and walk out with milk, a chocolate bar and a magazine. Ask yourself afterwards what the shelf looked like and you could not draw it. You did not read forty labels. You did what everyone does: looked at a fraction of what was there, went to the familiar, glanced at what other people were taking, and decided in a few seconds. Consumer behavior is the study of how that actually happens, as opposed to how a spreadsheet says it should. It sits where marketing meets psychology. Its famous results fill popular books, and a good share of them have not survived careful checking. Part of what a graduate learns is how to tell which is which. The course in one box Who takes itMarketing majors, usually in year 2, straight after the introductory marketing course (it is the second course in the Wharton marketing concentration). Psychology, design and business students take it as an elective. What it is forPredicting what real people notice, prefer and do, and what they will regret, so that products, shelves and messages fit them. Comes afterMarketing Management (34/57) and Behavioral Economics (34/46). Leads toBrand Management (34/60), Pricing (34/61), and, through Marketing Management, Marketing Research & Analytics (34/59). What this piece gives youYou won't be able to run a consumer study. You will know how little a shopper takes in, why crowds amplify luck, and which famous claims to doubt. Easier first reads: The Biases That Fool Everyone (17/02). This one goes further and checks the claims. You notice far less than you think Sharp vision covers only a small patch at the centre of your gaze. The rest of the scene is filled in by expectation. That is why people can look straight at something and not see it. In the best-known demonstration, Daniel Simons and Christopher Chabris asked viewers to count basketball passes in a video; about half of them did not notice a person in a gorilla suit walk through the middle (Perception, 1999). The effect has been repeated many times. The lesson for marketing is blunt: attention is the scarce resource, and the shopper is not looking where you hope. A cousin is change blindness. If a change happens during a blink, a cut or a flicker, people miss large differences. Ronald Rensink and colleagues built a "flicker" method in 1997 that shows how hard it is to spot a change when a blank screen separates two versions. Try it: one item on the shelf changes colour. Find it. Interactive · Spot the change on the shelf Start a new round Blank between frames: on Show me Press Start. Tap the item you think is changing. The shelf flickers about twice a second. Illustrative demo. Frames alternate: original, (blank), changed, (blank). With the blank off, the change makes a visible jump and is much easier to see. With the blank on, most people take several cycles; with it off, the moving colour jumps out. A real shelf has hundreds of items and no obvious blank, but every blink, glance and pause plays that role. For brands this means a distinctive colour or shape is not decoration: it is how the product gets noticed at all (see Brand Management, 34/60). Learning to want, and the decision journey People are not born wanting a brand. They learn to. Three mechanisms matter. Repetition: Robert Zajonc showed in 1968 that merely seeing something again makes people like it a bit more, and the "mere exposure" effect has held up in a large meta-analysis by Robert Bornstein (1989), with the effect modest and fading if repetition is overdone. Association: pair a brand with a feeling people already like (a sunny beach, a trusted face) and the feeling rubs off, a form of conditioning. Memory: you can only choose what comes to mind. Researchers call the handful of options a shopper actually thinks about the consideration set. Then comes the decision journey. The textbook picture is a funnel; in 2009 McKinsey consultants (Court, Elzinga, Mulder and Vetvik) argued it is a loop with four moments: initial consideration, active evaluation, purchase, and the experience after, which feeds back into the next consideration. For many everyday goods the journey is nearly empty: you grab the usual. The more a purchase costs, or the more risky it is, the longer and more deliberate the journey becomes. A graduate asks first how much of a journey this product has at all. MomentWhat happensWhat it implies ConsiderationA few options come to mindIf you aren't recalled, you aren't chosen EvaluationComparing, asking, searchingReviews and friends matter most here PurchaseOften fast, at the shelf or screenAvailability and defaults decide AfterUse, satisfaction or doubtFeeds the next round Choice: is more always worse? The most-quoted study in the subject is the "jam study" of Sheena Iyengar and Mark Lepper (2000). In a supermarket, a table with 24 jams drew more people to stop than a table with 6, but about 3% of the people at the big table bought, against about 30% at the small one. The conclusion, that too much choice paralyses, became a bestseller idea (Barry Schwartz, The Paradox of Choice, 2004). Then researchers tried to repeat it. In 2010 Benjamin Scheibehenne, Rainer Greifeneder and Peter Todd pooled 50 experiments and found that the average effect of more options was close to zero, with large differences between studies. Alexander Chernev and colleagues (2015) ran their own meta-analysis and argued the effect is real under conditions: when the options are hard to compare, when the shopper has no clear preference, and when the set is large and complex. So the honest reading: choice overload happens, but it is not a law, and the famous 3% versus 30% was not a typical size. A simple model shows why both can be true. Suppose every option has a random value for you between 0 and 1. The best of n options has expected value n ÷ (n + 1): more options always raise the best you could find, with diminishing returns. But looking costs effort. Subtract a small cost per option and a sweet spot appears. net = n ÷ (n + 1) − c × nn = options · c = effort cost of each option as a share of full satisfaction Interactive · How many options is best? Cost of looking at each option Options on the shelf A toy model, not data. It shows the shape of the argument: more options help up to a point, effort costs rise linearly, and the peak depends on how costly looking is for this person and this product. At half a point per option the best number is 13, but the curve is flat near the top, so a shelf with 6 or with 24 is within a few points of 13. The model does not say the jam study was wrong. It says the effect depends on effort, which is why big, confusing sets can hurt while familiar categories (milk, toothpaste) usually do not. Other people decide with us: reference groups and social proof A reference group is anyone whose view you use as a yardstick: family, colleagues, strangers who look like you. Robert Cialdini called the habit of using other people's behaviour as evidence social proof (Influence, 1984). It is sensible: if a restaurant is full, it is probably good. The risk is that proof can be circular, with each person copying the last. Matthew Salganik, Peter Dodds and Duncan Watts tested this in 2006. 14,341 participants downloaded unknown songs from an online "music market". Some saw others' download counts; some did not. When people could see the counts, the popular songs got much more popular, the unpopular ones less, and which songs won differed from one run of the experiment to the next. Quality mattered a little (the best songs rarely did badly and the worst rarely did well), but luck early on was amplified. The next model captures the mechanism: 200 people choose in turn between two equally good options. Interactive · Twenty parallel worlds of 200 choosers How much people copy the crowd so far Run again A toy model in the spirit of the 2006 study (not its data). Each person picks A with probability (1 − w) × ½ + w × the share of A so far. Both options are identical; each dot is one world's final share of A. With no copying, every world ends near 50-50. Turn the influence up and the worlds spread out: in some, A takes over; in others, B, though the two options are identical. Success looks like quality from inside any single world. This is why marketers pay for early reviews, why "bestseller" labels work, and why a hit can be partly an accident. The same ad, a different country Consumers carry a culture. Geert Hofstede's 1980 study, based on IBM employee surveys in more than 50 countries (about 117,000 matched samples, 1967–73), grouped national differences along dimensions such as individualism versus collectivism, power distance and uncertainty avoidance. In his published scores, the United States rates about 91 on individualism, while countries like Hong Kong sit near 25. Marketers used this to guess whether an ad should talk about "you" or about "family". Treat it as a hypothesis, not a map. Critics such as Brendan McSweeney (2002) challenge the method: the sample was one company and mostly men, countries are not uniform, a national average says little about any individual, and the data is old. A useful rule: use culture to ask a question ("would this joke, colour or family role mean the same here?") and then test it with local customers. Many tales about product names that "meant something rude" abroad are urban legends, so check before repeating one. Famous claims, checked The field has a pile of results that everyone has heard. Some were real, some shrank, some were never what they were said to be. Guess first; then see. Interactive · Held up, shrank, or never real? Claim 1 of 7 Held up Shrank or depends Not real as told Next claim Notice how often the number in the story is the part that fails, while a smaller, real effect remains. Two lessons for working marketers: ask for the sample and whether anyone repeated it, and be wary of any single "X% of decisions" figure. After the purchase: regret and doubt Leon Festinger's cognitive dissonance (1957) explains why people, after a costly purchase, seek reasons they chose well and avoid evidence they didn't: the mind dislikes holding "I chose it" and "it was a mistake" together. Sellers use this with follow-up messages that confirm the choice. Regret has a time pattern too. Thomas Gilovich and Victoria Medvec (1995) found that in the short run people regret what they did, but looking back over a long time they regret what they did not do. On the shelf, regret and choice interact. Barry Schwartz and colleagues (2002) found that "maximisers", who check everything for the best, report less satisfaction and more regret than "satisficers", who take what is good enough; the work is correlational and later studies are mixed, so treat it as a tendency, not a verdict. Returns policies, free trials and reviews are all ways of lowering the price of regret. What a graduate just sees The shopper looks at very little; the first job of a pack or a page is to be noticed. Most everyday buying is quick and habitual; the "journey" is short for cheap, familiar things. Repetition and association build liking before any argument is heard. More choice is not automatically worse; it depends on effort and on whether people know what they want. Crowds amplify luck: popularity is partly a story about earlier popularity. Culture is a question to test locally, not a stereotype to assume. When a study has a dramatic number, check whether it was repeated. People regret, and sellers can lower that cost by making returns and reassurance easy. Where it shows up next Brand Management (34/60): mental availability is the consideration set, and distinctive assets are the answer to attention. Pricing (34/61): how people perceive a price, from anchors to "is this fair". Marketing Research & Analytics (34/59): how to ask, test and check whether a finding replicates. Behavioral Economics (34/46): the formal sister of this subject, with the same replication debates. In the world: why retail shelves are laid out as they are, why apps rank "most popular", and why returns policies exist. If you want to go deeper 📘 Textbook. Michael Solomon, Consumer Behavior: Buying, Having, and Being (Pearson). The standard undergraduate text, with many applied cases. 📘 Textbook. Wayne Hoyer, Deborah MacInnis and Rik Pieters, Consumer Behavior (Cengage). More research-focused. 📘 Gentler. Robert Cialdini, Influence (1984; revised editions). Readable and full of examples. Daniel Kahneman, Thinking, Fast and Slow (2011) is the companion, though read it knowing some of its cited priming studies have been questioned. 🌐 Free course. MIT OpenCourseWare 15.810 Marketing Management covers the consumer side in its middle sessions; there is no single free consumer-behavior course I could confirm, so pair it with the books. The takeaway People notice little, copy others, decide fast and regret in patterns. Design for the sliver of attention you will get, remember that popularity feeds itself, and check any famous claim for a replication before relying on it. Try this: next time you shop, notice the first three things you looked at and the last thing you picked. Then ask which was seen first, and which came from habit. Sources & Further Reading Sources. Simons & Chabris, "Gorillas in our midst", Perception (1999). Rensink, O'Regan & Clark (1997), flicker method. Zajonc (1968); Bornstein (1989). Court et al., "The consumer decision journey", McKinsey Quarterly (2009). Iyengar & Lepper, JPSP (2000); Scheibehenne, Greifeneder & Todd, Journal of Consumer Research (2010: 50 experiments, mean effect near zero); Chernev, Böckenholt & Goodman, Journal of Consumer Psychology (2015). Salganik, Dodds & Watts, Science (2006; 14,341 participants). Hofstede (1980; 117,000 IBM samples, via Wikipedia summary); McSweeney (2002). James Vicary's 1957 popcorn claim (57.5% and 18.1% increases) was called "a gimmick" by Vicary in a 1962 Advertising Age interview. Festinger (1957); Gilovich & Medvec (1995); Schwartz et al. (2002). Maslow's pyramid: Bridgman, Cummings & Ballard, "Who built Maslow's pyramid?", Academy of Management Learning & Education (2019). The simulations are toy models; the numbers (best n = 13 at 0.5% cost; herding spread) were computed in node. This is not business advice. 🌐 MIT OCW 15.810 Marketing Management 📘 Solomon, Consumer Behavior 📘 Cialdini, Influence Step 14. Machine Learning: Programs That Write Their Own Rules -------------------------------------------------------------- Why: Machines that learn, and what they say about minds. Nobody wrote the rule that tells a spam message from your mother's email. A program found it by looking at examples, and everything in the field is about whether you can trust what it found. Your phone groups your photos by person. Your inbox hides the lottery scams. A bank flags a card payment at 3 a.m. from another continent. Try to write the rules for any of these by hand and you hit a wall within an hour: what, exactly, makes a face a face, or a message spam? Machine learning gives up on writing the rules. Instead you show a program thousands of labelled examples and let it adjust itself until it gets them right. This is powerful, and it quietly changes the questions you have to ask. The hard part is no longer "what is the rule?" but "how do I know the rule it found will work tomorrow, on people it has never seen?" That question, and a handful of simple tools for answering it, are what a graduate of a machine-learning course keeps for life. The course in one box Who takes it: Computer Science majors, usually in year 4; also statistics, data science and engineering students. What it's for: building programs that learn their behaviour from data, and judging honestly whether they work. Comes after: Linear Algebra (34/04), Probability & Statistics (34/05), Multivariable Calculus (34/02) and Algorithms (34/24). Leads to: no later subject in this series lists it as a prerequisite; it feeds research and careers in AI, and touches Computer Security (34/33), where learning systems are both tools and targets, and Databases (34/28), where the training data lives. Honest line: you won't train a modern language model after this; you will understand the training loop all of them share. For the AI story and a tour of language models, read Artificial Intelligence (29/01) and How Large Language Models Actually Work (29/02) first; this piece goes underneath them. Learning from data, and the data you hold back Take a toy problem. Each email gets a "suspicious words" score, and you must decide spam or not. You could write a rule by hand: spam if the score is above 3. But why 3? Better: try every threshold on a set of labelled examples and keep the one that gets the most right. That, in miniature, is training: a rule with an adjustable knob, a way to score the rule on data, and a search for the best setting of the knob. Here is the trap. The setting that fits the examples you used will almost always look better on those examples than on fresh ones, because you tuned it to them. So practitioners split their data three ways: Training set: the examples the model adjusts itself on. Validation set: held back to compare choices (which model, which settings). Test set: touched once, at the very end, to estimate how it will do in the real world. If you keep tuning against it, it stops being a test. Try it: pick the threshold, then meet fresh data Filled dots are 30 training emails (orange = spam, blue = not spam); hollow squares are 20 held-back ones the rule never saw. All the data are made up for this demo. The best threshold on the training emails is 3.3, which gets 93% of them right, but only 85% of the 20 held-back emails. Threshold: spam if score > 3.0 Let the data choose Linear regression and gradient descent Many problems are about predicting a number. The simplest model is a straight line: prediction = w × input + b. The two knobs are the slope w and the intercept b. To score a line, measure how far it misses each point, square the misses so that over and under both count, and average them: the mean squared error (MSE). The method of least squares was first published by Adrien-Marie Legendre in 1805; Carl Friedrich Gauss claimed to have used it earlier, which started a priority dispute. For a line there is a formula for the best w and b. For huge models there is not, and we use the idea that carries the whole field: gradient descent. Stand on the error surface, feel which way is steepest downhill (the gradient, from Multivariable Calculus 34/02), and take a small step that way. Repeat. Augustin-Louis Cauchy suggested the method in 1847. new w = w − η × (slope of the error with respect to w)    same for bη ("eta") is the learning rate, the size of each step The learning rate is a real trade-off: too small and you crawl; too large and you overshoot the valley and bounce out of it. Try it. The 12 points below are made up around a line; the least-squares answer is w = 2.465, b = 8.289, with MSE 2.58. Try it: fit a line by rolling downhill Start from w = 0, b = 0 and press Step. Choose a learning rate. With η = 0.01, 200 steps reach w = 2.89, b = 5.79 and MSE 4.37, still on the way down; with η = 0.04 the steps are too big and the error explodes instead of shrinking. η = 0.001 η = 0.01 η = 0.03 η = 0.04 StepRun 20 stepsReset Overfitting and underfitting A line is a rigid model. A more flexible one, say a polynomial with more bends, can follow the data more closely. Flexibility is a dial, and both ends of it fail: Underfitting: the model is too simple to capture the pattern. It is wrong on the training data and wrong on new data. Overfitting: the model is so flexible that it memorises the noise in the training data. It is nearly perfect there and bad on new data. This is why the validation set exists. Training error only ever goes down as you add flexibility. Validation error goes down, then comes back up, and the bottom of that U is where you stop. Training error tells you how well the model remembers. Validation error tells you how well it learned. Try it: turn the complexity dial The true pattern is a smooth wave; the 12 filled dots are noisy samples of it, and the 40 hollow dots are fresh noisy samples used only to judge. The curve is a polynomial fitted to the filled dots. A degree-1 line misses the wave (errors 0.54 and 0.43), degree 3 does best on fresh data (0.28 train, 0.35 validation), degree 9 hugs the 12 dots (0.14) but is far worse on fresh ones (1.02). Polynomial degree: 3 Classification and neural networks as layered functions Classification predicts a label rather than a number: spam or not, which digit, which animal. A simple classifier computes a weighted sum of the inputs and squashes it into a probability between 0 and 1, and its decision boundary is a straight line (or plane). That is enough when the classes can be separated by a line. Some problems cannot. The classic example is XOR: the answer is 1 when exactly one of the two inputs is 1. No single straight line separates the four cases correctly. In the late 1960s Marvin Minsky and Seymour Papert's book Perceptrons analysed exactly this kind of limit for single-layer networks, and is often credited with helping to cool interest in neural networks for years (historians debate how much). The escape is to stack models. A neural network is a chain of simple functions: each layer takes a weighted sum of its inputs, applies a bend (a non-linear function such as tanh), and hands the result to the next layer. One layer draws a line; a layer built on top of lines can carve out regions that no single line can. Training a network means finding good weights, and the tool is backpropagation, which in words goes: Run an example forward through the layers and get a prediction. Measure how wrong it was. Walk backwards, layer by layer, asking each weight "how much of this error was your fault?", using the chain rule from calculus. Nudge every weight a little in the direction that reduces the error. Repeat on the next example. The idea was derived several times; the 1986 Nature paper by David Rumelhart, Geoffrey Hinton and Ronald Williams is the one that made it popular. Gradient descent plus backpropagation, run at scale, is still how neural networks are trained, including the large language models in 29/02. Try it: teach a network XOR Four training cases. Choose how many hidden units the network has, then slide through training steps (always the same starting weights). The picture shows the network's output over the plane. With no hidden layer, the best the model can do is output about 0.5 for every case (loss 0.693) and it stays stuck; with 4 hidden units, 500 steps already get all four right (loss 0.010). No hidden layer 2 hidden units 4 hidden units Training steps: 500 Evaluation: accuracy is not enough "The model is 99% accurate" can be meaningless. Imagine a rare disease affecting 1% of 10,000 people. A model that always says "healthy" is 99% accurate and finds nobody. Count outcomes instead: true positives (TP), false positives (FP), true negatives (TN), false negatives (FN). precision = TP / (TP + FP): of those flagged, how many were rightrecall = TP / (TP + FN): of those who were sick, how many were found The two pull against each other. A model that flags only when it is certain has high precision and low recall; one that flags at the slightest hint has high recall and swamps you with false alarms. Where to set the threshold depends on what each error costs: missing a cancer is not like sending a good email to the spam folder. And when the real condition is rare, even a decent test produces mostly false alarms. Try it: a test with a threshold 10,000 people. The test gives a score; sick people score on average 2 standard deviations higher than healthy ones (a decent test, invented for this demo). Move the threshold and the prevalence. At 1% prevalence and a threshold of 1.0, the test finds 84 of 100 sick people (recall 84%) but raises 1,571 false alarms, so only 5% of those flagged are actually sick. 1% are sick 10% 50% Flag when score > 1.0 Bias: the model learns the data you gave it A model has no opinion about fairness. It finds whatever pattern best fits the examples it was shown, including patterns in who is in the examples. If one group is a small part of the training data, the model's single rule is tuned to the majority and does worse on the rest. This is not hypothetical. In the 2018 Gender Shades study, Joy Buolamwini and Timnit Gebru found that the two benchmark face datasets they examined were 79.6% (IJB-A) and 86.2% (Adience) lighter-skinned subjects, and that three commercial gender-classification systems misclassified darker-skinned women at error rates of up to 34.7%, against at most 0.8% for lighter-skinned men. Try it: who is in the training data? Two groups, A and B, whose scores sit on different scales (an invented example). The model picks the one threshold that gets the most training cases right. Change how much of the training data is group B. With 2% of the data from group B the model is 84% accurate for A but 74% for B; at 50% both groups get 81%. Group B share of training data: 2% More data from the same source does not fix this; different data, or a model checked group by group, can. Bias can also come from the labels (past human decisions the model copies) and from how a system is used. Checking performance for each group, not just on average, is the habit to keep. The same discipline applies to the newest systems. The Stanford HAI AI Index 2026 report describes a "jagged frontier": one model earned a gold medal at the International Mathematical Olympiad, yet the top model it tested read analog clocks correctly only 50.1% of the time. An average score, or a headline result, says little about how a model will do on the case in front of you; that has to be measured. What a graduate just sees Learning means tuning knobs to fit examples; the rule is found, not written. Performance on the data you trained on is not evidence. Only data held back is. Gradient descent is "step downhill"; the learning rate is the step size and it matters. Too simple and too flexible both fail; validation error has a U shape. A network is stacked simple functions, and backpropagation is the chain rule applied layer by layer. Accuracy hides rare cases; ask for precision, recall and the cost of each mistake. When a thing is rare, most alarms are false alarms. A model is only as fair as its data; check each group, not the average. Where it shows up next Linear Algebra (34/04) and Multivariable Calculus (34/02): weights are vectors and matrices, and the gradients are partial derivatives. Probability & Statistics (34/05): the train/test idea is the statistician's out-of-sample check, and precision is Bayes' theorem in disguise. Algorithms (34/24): training is optimisation, and its cost is why hardware like GPUs matters (see 29/12). Computer Security (34/33): learning systems can be fooled by crafted inputs and poisoned data. In the world: fraud screening, search ranking, recommendations, translation, medical imaging, and the chatbots in 29/02. As of the 2025 Stanford AI Index, the compute used to train notable models was reported to be doubling roughly every five months, which is why training is now an industrial activity. If you want to go deeper 📘 An Introduction to Statistical Learning (James, Witten, Hastie, Tibshirani; 2nd edition 2021, with a Python edition in 2023): free from the authors, the standard first university text; clear and light on maths. 📘 Deep Learning (Goodfellow, Bengio, Courville, MIT Press 2016): the standard graduate text on neural networks, free to read online. ▶️ Gentler: 3Blue1Brown's neural-networks video series builds the ideas above with pictures. 🌐 Free courses: MIT OpenCourseWare 6.036 Introduction to Machine Learning and Stanford CS229. The takeaway Machine learning replaces hand-written rules with knobs tuned to examples, using gradient descent to find the settings. Everything else is about trust: hold data back to measure honestly, find the middle between too simple and too flexible, stack functions when a line is not enough, ask which errors matter, and check who is in the data. Try this: the next time you read "an AI that is 95% accurate", ask four questions. Accurate on what data? How rare is the thing it looks for? What does each kind of mistake cost? And has anyone checked it for the group you belong to? Sources: A.-M. Legendre (1805) and C. F. Gauss on least squares (priority dispute, as summarised in standard histories of statistics); A.-L. Cauchy (1847) on gradient descent; M. Minsky and S. Papert, Perceptrons, MIT Press, 1969; D. Rumelhart, G. Hinton and R. Williams, "Learning representations by back-propagating errors", Nature 323, 1986; J. Buolamwini and T. Gebru, "Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification", Proceedings of Machine Learning Research 81, 2018; Stanford HAI, AI Index Report 2025 (training compute doubling about every five months) and AI Index Report 2026 (the "jagged frontier": IMO gold, 50.1% on analog clocks); G. James, D. Witten, T. Hastie, R. Tibshirani, An Introduction to Statistical Learning. All datasets in the demos are invented, and every number they show was computed and cross-checked for this piece. Go deeper: 📘 An Introduction to Statistical Learning (statlearning.com, free) 📘 Deep Learning — Goodfellow, Bengio, Courville (deeplearningbook.org, free online) ▶️ 3Blue1Brown: Neural Networks 🌐 MIT OCW 6.036 Introduction to Machine Learning