Economics
A path through The Default Library
========================================
How people, firms and countries choose when they can't have everything, and how to tell what actually caused what.


Step 1. Calculus I: Slopes, Totals and the Bridge Between Them
--------------------------------------------------------------
Why: Margins are derivatives: economics is the study of "one more".
Your car's speedometer is a derivative. Its odometer is an integral. Everything a first calculus course teaches is a way of moving between the two.

Look at a dashboard. The odometer says how far you have gone in total. The speedometer says how fast the total is changing right now. Press the accelerator and the speedometer moves first; the odometer follows, a little behind, like a patient accountant adding up what the speedometer reported.
  Nobody gave the speedometer a ruler or a clock to look at the whole trip. It reads one instant. Yet it reports a rate, which is a statement about change, from a single moment. How can a single moment say anything about change? That puzzle is what the course exists to answer, and the answer turned out to run most of modern engineering and computing.
  This is the second time the library has met calculus. Calculus: Derivatives, Integrals, and the Language of Motion (12/11) is the easier first read, with the tangent line and the rectangle sum. Here we go a step further: what the course trains you to see, how a computer gets it wrong, and where the idea is used afterwards.

  
    The course in one box
    Who takes it: first-year students in Mechanical & Aerospace Engineering and in Computer Science, and almost every other science or engineering degree.
    What it is for: turning "how much?" into "how fast?" and back, so that change can be calculated instead of guessed.
    Comes after: nothing in this series. Algebra and a little trigonometry are all it assumes.
    Leads to: Multivariable Calculus (34/02), Differential Equations (34/03), Probability & Statistics (34/05), Physics I: Mechanics (34/06), Thermodynamics (34/14), and from them nearly everything else on both majors.
    Honest line: you will not be able to differentiate a hard formula by hand after this. You will see why a flat spot on a curve is where the best answer hides, and why the area under a rate is a total.
  

  A limit is where a calculation settles
  The speedometer puzzle is this. Speed is distance divided by time, but in an instant no time passes and no distance is covered, so you seem to be dividing zero by zero. The way out is to stop asking what happens at zero and ask what a sequence of better and better answers is settling towards. Measure over one second, then a tenth, then a hundredth, and watch the average speed home in on a number. That number is the limit.
  speed now = limit of (distance change ÷ time change) as the time change shrinks to 0
  A graduate does not carry the formal definition around, but carries the habit: if a quantity looks like 0 ÷ 0, ask what it is approaching. Take sin(x) ÷ x as x goes to zero, with x in radians. At x = 0.1 it is 0.99833; at 0.01, 0.99998; at 0.001, 0.9999998. It never "is" 1 at zero, since it is undefined there, but it settles on 1. That one limit is why the derivative of sine turns out to be cosine.

  
    Interactive · Shrink the gap until the computer gives up
    We estimate the slope of ex at x = 1 by (e1+h − e) ÷ h. The true slope is e = 2.718281828459045. Make the gap h smaller.
    Gap h = 10−k
      
    
    
  
  The error falls as the gap shrinks, as the limit promises, until about h = 10−8, where it bottoms out near 7 × 10−9. Shrink further and it gets worse: the two values being subtracted agree in almost all of their roughly 16 stored digits, so the subtraction wipes out the information. At 10−16 the answer is exactly 0. This is a computer-science lesson hiding in a calculus course: the limit is a mathematical idea, and a machine can only approach it so far.

  The derivative is a slope, and a slope is a rate
  Zoom in on any smooth curve and it straightens out. The steepness of that straight piece is the derivative. In plain words: how much the output moves per unit nudge of the input. If the curve is distance against time, the derivative is speed. If it is temperature against position, it is how sharply it warms as you walk. Rate and slope are one thing seen two ways.
  f′(x) = limit of [f(x+h) − f(x)] ÷ h as h → 0
  The graduate's real asset is not computing derivatives but reading them. Where the curve is rising, the derivative is positive. Where it is flat, the derivative is zero. Where the curve is steepest, the derivative is at its peak. Try it: the lower graph is the derivative of the upper one, drawn point by point.

  
    Interactive · A curve and its slope, side by side
    
      x³ − 3x
      sin x
      bell: e^(−x²)
    
    Position x
      
    
    
  
  Check the three curves. Every time the top curve peaks or dips, the bottom one crosses zero. Every time the top is climbing, the bottom is above the axis. For x³ − 3x the slope is 3x² − 3, which is zero at x = −1 and x = 1: the hill top and the valley floor.

  The chain rule: rates multiply down a chain
  Rates often come in chains. A bicycle's pedals turn the chainring, the chain turns the rear wheel, the wheel moves the bike. If the wheel turns 3 times per pedal turn, and the bike moves 2 metres per wheel turn, the bike moves 6 metres per pedal turn. You multiply the ratios. That is the chain rule, and it is the most used rule in the course.
  d(outer)/d(input) = d(outer)/d(middle) × d(middle)/d(input)
  The classic worked case is a balloon. Its volume depends on its radius, and the radius depends on time. At radius r the surface has area 4πr², so one extra centimetre of radius adds a shell of volume about 4πr² cm³: that is dV/dr. Multiply by how fast the radius is growing and you have how fast the volume is growing. At r = 10 cm, growing 2 cm each second, volume grows at 4π × 100 × 2 ≈ 2,513 cm³ per second. Pull the sliders and confirm it.

  
    Interactive · Inflate a balloon
    Radius now
      
    Radius is growing at
      
    
    
  
  The picture shows why it works: the shaded shell is what one second adds, a skin of area 4πr² and thickness equal to the radius growth. The same rule runs inside every neural network, where the "chain" is dozens of layers long and the multiplication is called backpropagation (Machine Learning, 34/32; Multivariable Calculus, 34/02).

  Where the slope is zero, something is best
  At the top of a hill you stop going up and have not yet started going down. The slope is zero. So to find the biggest or smallest value of anything smooth (strongest, cheapest, lightest, fastest) look for where its derivative is zero. That is optimisation, and it is why engineers love the derivative.
  Take a 30 cm by 20 cm sheet of card. Cut a square of side x from each corner and fold up the sides to make an open box. Its volume is V = x(30 − 2x)(20 − 2x). Cut too little and the box is shallow. Cut too much and it is narrow. In between there is a best x, and the slope tells you where: at x ≈ 3.92 cm the volume peaks at about 1,056 cm³.

  
    Interactive · The best box
    Corner cut x
      
    Jump to the flat spot
    
    
  
  Notice what the slope does as you drag. To the left of the peak the slope is positive, so a bigger cut helps. To the right it is negative, so a smaller cut helps. At the peak it is zero and nothing nearby is better. Gradient descent, the algorithm that trains machine-learning models, is exactly this reasoning repeated: look at the slope, step downhill, repeat.

  The integral adds up a rate into a total
  The odometer is the speedometer added up over time. Draw speed against time; the area between the curve and the time axis is distance. That is the integral. Anywhere a rate is known and a total is wanted (litres per minute into a tank, watts into a battery, newtons over a distance) the answer is an area.
  Try it before you read on. A car accelerates steadily from 0 to 20 m/s in 10 seconds, cruises at 20 m/s for 30 seconds, then brakes steadily to a stop in 10 seconds. How far has it gone?

  
    Interactive · Predict, then reveal
    
      600 m
      800 m
      1,000 m
      1,200 m
    
    
    Choose an answer.
  
  The shape is two triangles and a rectangle: ½ × 10 × 20 + 30 × 20 + ½ × 10 × 20 = 100 + 600 + 100 = 800 m. The common wrong guess, 1,000 m, uses the top speed for all 50 seconds and forgets that the car spends time slower than that.

  The Fundamental Theorem: two jobs, one idea
  Here is the surprise that makes the course a single subject. Build a running total from a rate: the amount in the tank after time T. Then ask how fast that total is growing at T. The answer is just the rate at T. Accumulating and differentiating undo each other, the way the odometer and speedometer do. Newton and Leibniz, working separately in the late 1600s, each saw this.
  if A(T) = area under the rate up to T, then A′(T) = the rate at T
  Its practical power is that you can find a total without adding anything: find a function whose slope is your rate and subtract its values at the two ends. Below, water flows into a tank at a rate that wobbles: 4 + 3 sin(t/2) litres per minute. Drag the time and watch the right-hand graph, the tank's total, steepen exactly when the rate on the left is high.

  
    Interactive · The tank and its rate
    Time T
      
    
    
  
  The readout checks the theorem two ways: area by adding 600 thin slices, and the exact formula 4T + 6(1 − cos(T/2)). At T = 12 minutes both give 48.24 litres. The slope of the total at that moment is 4 + 3 sin 6 ≈ 3.16 litres per minute, the rate at that moment.

  Every smooth thing is nearly a straight line
  The last habit is the most useful in daily engineering. Close to any point, a smooth curve is almost its tangent line, so you can replace it with a straight line and be nearly right. The small-angle rule sin x ≈ x is why a pendulum swings at a steady rhythm. The rule √(1 + x) ≈ 1 + x/2 gives mental square roots: √1.1 ≈ 1.05, true value 1.0488.
  f(a + x) ≈ f(a) + f′(a) · x   (good when x is small)
  How small is small? The error shrinks faster than the step. Halve the step and the error falls about four times for √(1 + x) and ex, and about eight times for sine.

  
    Interactive · How good is the straight line?
    
      sin x ≈ x
      √(1+x) ≈ 1 + x/2
      eˣ ≈ 1 + x
    
    Step x from the point
      
    
    
  

  What a graduate just sees
  
    A rate and a slope are the same thing, and "how fast?" always means a derivative.
    A total is an area under a rate, and adding up thin slices is a calculation, not a mystery.
    Rates multiply along a chain: gear ratios, units, layers of a network.
    The best value of a smooth quantity sits where its slope is zero.
    Differentiating and accumulating are inverse jobs, so a total can often be found without adding.
    Up close, everything smooth is a straight line, and engineers exploit that constantly.
    Subtracting two nearly equal numbers on a computer destroys precision, so a naive slope estimate gets worse when the gap gets too small.
  

  Where it shows up next
  
    Physics I (34/06) and Dynamics (34/12): velocity is the derivative of position, acceleration of velocity, and Newton's law ties force to the second one.
    Differential Equations (34/03): when a law is written as a rate, solving it means integrating, which is the whole next course.
    Thermodynamics (34/14) and Heat Transfer (34/16): heat added is an integral of power; temperature profiles are read from slopes.
    Probability & Statistics (34/05): the chance of a range of outcomes is the area under a density curve.
    Algorithms (34/24) and Machine Learning (34/32): growth rates of running time are compared with limits, and models are trained by following slopes downhill.
  

  If you want to go deeper
  
    📘 Textbook: James Stewart, Daniel Clegg and Saleem Watson, Calculus: Early Transcendentals (9th ed., Cengage). The most-assigned text; best for worked examples and a huge problem bank.
    📘 Textbook: Hass, Heil, Bogacki and Weir, Thomas' Calculus (15th ed., Pearson). Strong on applications, popular in engineering departments.
    ▶️ Gentler: Grant Sanderson's Essence of Calculus series on 3Blue1Brown, a visual tour of the same ideas. Also Steven Strogatz, Infinite Powers, for the story.
    🌐 Free course: MIT OpenCourseWare 18.01SC Single Variable Calculus, with video lectures, notes and problems. Khan Academy also covers it.
  

  
    The takeaway
    Calculus I teaches two moves and one bridge. The derivative turns a total into a rate (a slope); the integral turns a rate into a total (an area); and the Fundamental Theorem says they undo each other. From that come the habits of mind: chain rates together, look for zero slope to find the best, and treat smooth things as lines up close. Our slope estimate for ex was best near a gap of 10−8 and useless at 10−16, a reminder that the ideas are exact and the machines are not.
    Try this: next time you are in a car or on a bike, watch the speedometer and ask what the area under its history means. Then on a hill, find the spot where the road is flat. That is a derivative equal to zero.
  

  
    Sources & Further Reading
    Sources. Every number in the interactives was computed in node: the forward-difference error of ex at x = 1 for h = 10−1 to 10−16 (best 6.6 × 10−9 at 10−8, exactly 0 at 10−16, in IEEE double precision); sin(x)/x at 0.1, 0.01, 0.001; the box V(x) = 4x³ − 100x² + 600x, zero slope at x = (50 − √700)/6 = 3.9237, V = 1,056.3 cm³; the balloon 4π × 10² × 2 = 2,513.27 cm³/s; the car, 800 m; the tank, A(12) = 48.239 L; small-step errors of the three approximations. Book editions checked against the publishers' pages (Cengage, Pearson) and the course numbers against MIT OpenCourseWare.
    
      🌐 MIT OpenCourseWare 18.01SC Single Variable Calculus
      ▶️ 3Blue1Brown: Essence of Calculus
      🌐 Khan Academy: Calculus 1
      📘 Stewart, Clegg and Watson, Calculus: Early Transcendentals, 9th ed. (Cengage)
      📘 Hass, Heil, Bogacki and Weir, Thomas' Calculus, 15th ed. (Pearson)
      📘 Steven Strogatz, Infinite Powers, a popular history of calculus


Step 2. Microeconomics: Why Nothing Is Free and Every Price Is a Message
------------------------------------------------------------------------
Why: Supply, demand, incentives and when markets fail: the toolkit everything else uses.
Everything costs what you gave up, choices happen one step at a time, and a price is the signal that makes strangers cooperate. Until it is wrong, and then you can see exactly how wrong.

A coffee shop raises its flat white from 3.50 to 3.80 and, somehow, still looks busy. A ride-hailing app doubles its fares in the rain and, ten minutes later, there are more cars on the road. A city starts charging drivers to enter its centre, and the jam thins out without a single new rule about who may drive. Nobody commanded any of this. A price moved, and thousands of people each did a small, selfish, sensible thing.
  An economics graduate watches those three scenes and sees the same few ideas working: everything costs something you gave up, decisions happen one step at a time, and a price is a signal that tells buyers to hold back and sellers to show up. That toolkit is microeconomics: how individuals and firms choose, and what happens when those choices meet in a market. It is also the friendliest door into the whole subject, because you already use it every time you shop.

  
    The course in one box
    
      Who takes itEveryone on an economics, business or marketing degree, usually in year 1 as "principles" and again in year 2 as "intermediate microeconomics", with calculus. MIT puts its principles course (14.01) at the gateway to Course 14-1; business schools such as Wharton teach the same toolkit as "managerial economics".
      What it is forPredicting how people and firms respond when prices, taxes, rules or competitors change, and judging whether the result is good for anyone besides the people making the trade.
      Comes afterCalculus I (34/01): "one more" is a derivative.
      Leads toMacroeconomics (34/43), Game Theory (34/45), Behavioral Economics (34/46), International Trade & Finance (34/47), Public Economics (34/49), Strategy & Competition (34/54), Marketing Management (34/57) and Pricing (34/61).
      What this piece gives youYou won't be able to estimate a demand curve from real data. You will see why a price cap makes queues, why a tax costs more than it raises, and why a market with five firms behaves differently from one with one. Easier first reads: Where Money Actually Comes From (04/02), Pricing Without Fear (04/03) and Capitalism: How Markets, Prices and Private Property Organise an Economy (31/01). This one puts numbers and sliders on the same ground. All the markets below are illustrative, not data.
    
  

  Scarcity and opportunity cost: the real price is what you give up
  Wants have no ceiling; time, land, labour and money do. That gap is scarcity, and it makes every choice a trade. The honest price of anything is its opportunity cost: the best thing you gave up to get it. A two-hour film costs the ticket plus whatever else those two hours could have been. A university degree's biggest cost is often the wages you didn't earn.
  Economists draw this as a production possibilities frontier: the most of one good you can make for each amount of another. It bows outward because resources aren't equally good at everything. The first fields moved from wool to wheat are the ones best suited to wheat; the last fields are poor wheat land. So each extra unit costs more than the one before. This rising cost, not any law of nature, is what lets prices carry information.
  
    Interactive · A farm's frontier
    Wheat grown
      
    
    
    Illustrative farm: wool = 100 − wheat²/100. The cost of the next unit is wool(now) − wool(after one more wheat).
  
  At 40 wheat the farm can have 84 wool. One more wheat costs 0.81 wool. At 80 wheat the same extra unit costs 1.61 wool, twice as much. Nothing got worse; you just ran out of the easy fields. Remember this shape whenever someone says "just do more of it".

  Thinking at the margin: only the next unit matters
  People rarely choose all or nothing. They choose one more: one more hour of study, one more loaf baked, one more floor on the building. The margin is the edge where the decision lives. The rule an economics graduate carries: keep doing something while its marginal benefit is at least its marginal cost, and stop when the next unit would cost more than it brings in.
  do more while  marginal benefit ≥ marginal costfor a seller facing a market price, marginal benefit is just the price
  A bakery sells loaves at a fixed price. Its cost of the next loaf rises as the oven gets crowded and staff go onto overtime (an illustration, not a real bakery). The slider below is the owner's decision. Watch the profit hump: the top is exactly where the cost of the next loaf meets the price.
  
    Interactive · How many loaves?
    Price per loaf
      
    Loaves baked
      
    
    
    Illustrative: total cost = q + 0.05q², so the next loaf costs about 1 + 0.1q.
  
  At a price of 6 the best output is 50 loaves and profit is 125. At 40 loaves the owner is leaving money on the table; at 60 she is baking loaves that cost more than they sell for. A higher price pushes the best output up: at 10 it is 90. Lining up each price with its best output is exactly how a supply curve is built.

  Supply and demand: the price that clears the market
  Demand says that, other things equal, buyers want less as the price rises. Supply says that sellers offer more as it rises (the bakery just showed why). Draw both against price and they cross at one point: the equilibrium price, where the quantity buyers want equals the quantity sellers offer. Above it there is a surplus and sellers cut prices; below it there is a shortage and buyers bid up. The price is pulled to the crossing without anyone setting it.
  The key habit: separate moving along a curve (the price changed) from shifting it (something else changed: tastes, income, input costs, a frost). Try three shocks in a market whose starting point is price 30, quantity 40.
  
    Interactive · Shift the curves
    
      Reset
      Demand rises (a fad)
      Supply falls (a frost)
      Both
    
    Demand shift (units at every price)
      
    Supply shift (units at every price)
      
    
    
    Illustrative: demand Q = 100 − 2P, supply Q = 2P − 20, then shifted sideways.
  
  A fad lifts both price (35) and quantity (50). A frost lifts the price to the same 35 but cuts the quantity to 30. Same price, opposite quantities: a graduate reads the quantity to tell whether demand or supply moved. When both shift by 20 the price jumps to 40 and the quantity barely changes, because the two effects on quantity cancel.

  Elasticity: how loudly buyers react
  A price rise always lowers the quantity demanded. The question that matters is how much. Price elasticity of demand is the percentage fall in quantity for a 1% rise in price. If it is below 1 in size, demand is inelastic: buyers shrug (salt, insulin, petrol in the short run). If it is above 1, demand is elastic: buyers walk (one brand of cereal, a restaurant meal).
  elasticity = % change in quantity ÷ % change in pricesellers' revenue rises with price when demand is inelastic, and falls when it is elastic
  A real case: from January 2014 Mexico taxed sugary drinks at one peso a litre, commonly described as roughly a tenth of the price. Purchases of the taxed drinks fell about 6% over 2014, and about 9% in the lowest-income households (Colchero and colleagues, BMJ, 2016). If the whole tax reached shelf prices, that is an elasticity around 0.6: buyers responded, but less than one-for-one. It is a rough reading, since pass-through to prices varied.
  
    Interactive · What does a 10% price rise do?
    
      Staple (0.3)
      Unit elastic (1)
      Restaurant meal (2)
      One brand (3)
    
    Elasticity (size)
      
    Price
      
    
    
    Illustrative constant-elasticity demand: Q = 100 × (P ÷ 10)−e. Real elasticities differ by market, time horizon and how many substitutes exist.
  
  With elasticity 0.5, a 10% price rise cuts quantity by about 4.7% and raises revenue by about 4.9%. At 2, the same rise cuts quantity by 17.4% and revenue by 9.1%. That is why a monopolist never charges where demand is inelastic: it could earn more by charging more.

  Surplus and the triangle nobody gets
  A trade happens only if it makes both sides better off. Consumer surplus is the gap between what a buyer would have paid and what they did pay. Producer surplus is the gap between what a seller received and the least they would have taken. In the market above, each is worth 400, so the total is 800. A competitive market, left alone and free of side effects, produces exactly the trades that make that total as large as it can be.
  Now interfere. A tax drives a wedge between what buyers pay and sellers keep, so some trades that were worth doing no longer happen. A price ceiling below the equilibrium makes sellers supply less than buyers want. Either way the lost trades are the deadweight loss: surplus that went to nobody. It is a triangle, so it grows with the square of the wedge.
  
    Interactive · A tax, or a price cap
    
      Tax per unit
      Price ceiling
    
    Tax per unit
      
    
    
    Same market as above (price 30, quantity 40, surplus 800). For the cap, the best case is assumed: the buyers who value the good most get it. Real rationing is usually worse.
  
  A tax of 10 shrinks the quantity from 40 to 30, raises 300, and destroys 50 of surplus. Double the tax to 20 and the deadweight loss does not double, it quadruples to 200, while revenue rises only to 400. A ceiling at 20 leaves 20 units on sale where buyers want 60: a shortage of 40, a loss of 200 even in the best case, and an invitation to queue, bribe or build a black market. Note what a tax can do that a cap cannot: it raises money that can be spent. The next piece in this thread, Public Economics (34/49), is about who ends up actually paying it.

  Market structures: from a thousand sellers to one
  "Supply and demand" assumes sellers are too small to move the price. That holds for a farmer selling wheat, not for an electricity grid or a search engine. Economists line up a spectrum. At one end is perfect competition: many sellers, identical goods, so each takes the price as given and price is driven to marginal cost. At the other is monopoly: one seller, who can choose the price, and who restricts quantity to keep it high. In between are oligopoly (a few firms who watch each other) and monopolistic competition (many firms with slightly different products, like restaurants).
  The simplest model that shows the whole spectrum is Cournot's, from 1838: firms choose quantities, each knowing the others do the same. The more firms, the more each fears flooding the market, and the closer price falls to cost.
  
    Interactive · How many rivals?
    
      Firms compete on quantity
      Perfect competition
    
    Number of firms
      
    
    
    Illustrative: demand P = 100 − Q, each firm's marginal cost 20. One firm sets 40 units at 60. Cournot with n firms: Q = 80n ÷ (n+1).
  
  One firm charges 60 and sells 40; two firms get 46.7; five get 33.3; twenty get 23.8, nearly the competitive 20. Buyers gain as firms are added (consumer surplus goes from 800 with one firm to about 2,900 with twenty), and the deadweight loss shrinks from 800 to about 7. This is the economic case for competition policy, and the reason Strategy & Competition (34/54) is a whole course. Entry matters more than the headcount, though: a market with one firm but easy entry can behave better than a market with three that are protected.

  Externalities and public goods: when prices miss something
  Prices work when the buyer and seller bear all the costs and benefits of the trade. An externality is a cost or benefit that lands on someone else: smoke from a factory, a neighbour's loud party, or, on the good side, a beekeeper's bees pollinating the orchard next door. The market quantity is too high for harms and too low for benefits, because the decision-maker ignores the third party.
  The classic fix is a Pigouvian tax (A. C. Pigou, 1920): charge the polluter the damage they cause, so the private price includes the social cost. London's congestion charge, which began on 17 February 2003 at five pounds a day (the standard rate is now £18 as of January 2026), is a price on road space for exactly this reason. Ronald Coase argued in 1960 that, where bargaining is cheap and rights are clear, the parties can sort it out themselves with no tax at all. Both are alive in every environmental debate.
  
    Interactive · A factory that pollutes
    Damage per unit made, to neighbours
      
    
      No policy
      Tax equal to the damage
    
    
    
    Same market as before. Making each unit costs the neighbours the amount on the slider.
  
  With damage of 10 per unit, the market still makes 40 units but the best quantity is 30. The ten extra units each cost society more than they are worth: a loss of 50. A tax of 10 moves the market to exactly 30. The other kind of miss is a public good: something nobody can be excluded from and that one person's use doesn't use up (a lighthouse, clean air, a vaccine's herd effect). Each person would like someone else to pay, so private markets undersupply it. When a resource is shared but rivalrous, a commons, Garrett Hardin's 1968 "tragedy of the commons" warned of overuse. Elinor Ostrom's field work, in Governing the Commons (1990) and her 2009 Nobel Prize, showed that communities from Swiss pastures to fisheries often build rules that avoid it without privatising or handing it to the state. The honest version is "it depends on the rules". More on this in Public Economics (34/49).

  What a graduate just sees
  
    Every choice has an opportunity cost, and it usually rises: the first unit is cheap, the thousandth is not.
    Decisions happen at the margin. Sunk costs and averages are the wrong numbers to look at.
    A price is information. When it moves, ask which curve shifted, and read the quantity to tell.
    Whether a price rise helps a seller depends on elasticity, which depends on substitutes and time.
    A price cap makes a shortage and a tax makes a wedge; the deadweight loss grows with the square of the wedge.
    "More competitors" and "freer entry" push price toward cost; fewer push it away.
    When someone else bears the cost or gets the benefit, the market quantity is wrong, and the fix is to make the price carry it.
    Markets can fail, but "the government fixes it" is a claim to check too: compare real alternatives, not an ideal against a market.
  

  Where it shows up next
  
    Macroeconomics (34/43): the same supply-and-demand logic, scaled up to the price level, interest rates and the whole labour market.
    Game Theory (34/45): when the firm count is small, strategy replaces the price-taking assumption. The oligopoly dial above is a simple game.
    Public Economics (34/49): externalities, public goods and who bears a tax become the whole subject.
    In the world: a surge price on a ride app, a carbon price, a minimum wage debate, antitrust cases against big platforms, and every "should we charge for this?" meeting in a company.
  

  If you want to go deeper
  
    📘 Textbook. N. Gregory Mankiw, Principles of Microeconomics (Cengage). The most widely assigned introduction in the US and much of the world: clear, short chapters.
    📘 Textbook. Daron Acemoglu, David Laibson and John List, Economics (Pearson), or Paul Krugman and Robin Wells, Microeconomics (Worth). Two modern alternatives: the first leans on evidence and experiments.
    📘 Gentler. Charles Wheelan, Naked Economics (W. W. Norton). Almost no maths; the ideas as stories.
    🌐 Free course. MIT OpenCourseWare 14.01 Principles of Microeconomics (Fall 2023): lectures and problem sets. Also CORE Econ's The Economy and Khan Academy Microeconomics, both free.
  

  
    The takeaway
    Every choice costs the best alternative you gave up, and it is made one step at a time. Prices carry that cost between strangers: when they are right, markets turn scattered selfishness into trades that make both sides better off. When they are wrong (a cap, a tax, a monopolist, a neighbour who bears the smoke), something is lost, and the skill is to find which wedge is the problem.
    Try this: this week, price one thing you buy in terms of what you gave up, and find your margin. Would you pay for one more? Then watch a price move at a shop or an app and ask which curve shifted.
  

  
    Sources & Further Reading
    Sources. The ideas are standard textbook microeconomics (Mankiw; Acemoglu, Laibson and List). Every number in the interactives uses invented, illustrative markets and was computed in node: farm frontier 84 to 83.19 (cost 0.81) at 40 wheat and 36 to 34.39 (1.61) at 80; bakery best output 50 with profit 125 at price 6, 90 at price 10; equilibrium 30 and 40, shocks 35 and 50 (fad), 35 and 30 (frost), 40 and 40 (both); elasticity 0.5 gives −4.65% quantity and +4.88% revenue, elasticity 2 gives −17.36% and −9.09%; tax of 10 gives quantity 30, revenue 300, deadweight loss 50, tax 20 gives 200 and revenue 400; ceiling 20 gives quantity 20, shortage 40, loss 200; Cournot with 1, 2, 5, 20 firms gives price 60, 46.7, 33.3, 23.8 and consumer surplus 800 to about 2,900; externality of 10 gives optimum 30 and loss 50. Mexico's 2014 sugary-drink tax: Colchero, Popkin, Rivera and Ng, BMJ 352:h6704 (2016), which reports a 6% fall in taxed-drink purchases in 2014 and 9% in the lowest-income households; the one-peso-a-litre rate is as reported in public summaries and the "about a tenth of the price" is approximate. London congestion charge: began 17 February 2003 at £5, £18 from 2 January 2026 (Transport for London, via Wikipedia). Coase 1960 (Nobel 1991), Pigou 1920, Hardin 1968 (Science), Ostrom 1990 (Nobel 2009), Cournot 1838 are standard history of economic thought.
    
      🌐 MIT OCW 14.01 Principles of Microeconomics
      🌐 CORE Econ, The Economy
      📘 Mankiw, Principles of Microeconomics
      📘 Wheelan, Naked Economics


Step 3. Macroeconomics: The Economy as One Machine, and Why It Stalls
---------------------------------------------------------------------
Why: Growth, inflation, unemployment and central banks: the economy as a whole.
Output that compounds, prices that drift, jobs that flow, banks that create money, central banks with one lever. And four schools that disagree on how hard to push it.

In June 2022 a US shopper's costs were about 9% higher than a year earlier; in October 2022 the euro area peaked at 10.6%; in Turkey the annual average that year was 72%, and in Argentina in 2024 it was 220%. Meanwhile in Japan prices barely moved for a generation, and in Spain in 2013 more than one worker in four had no job. Same planet, same technology, wildly different experiences. The questions behind them are the ones macroeconomics asks: why does an economy grow, why do prices rise, who gets work, and what, if anything, can a government or a central bank do about it?
  A macroeconomics graduate reads those headlines as flows through a few connected pipes: output, prices, jobs, money and debt. The pipes are simple; the arguments over how to turn the taps are not, and this piece tells you who argues what. The data below is real and dated, from several regions. The models in the sliders are textbook toys with invented numbers, labelled as such.

  
    The course in one box
    
      Who takes itEconomics, business and public-policy students, usually in year 1 or 2 as "principles of macroeconomics" (MIT's is 14.02), then again in year 3 as "intermediate macro" with calculus. Business schools teach a shorter "global economy" version in the first year: Harvard Business School's required first-year curriculum includes the global economy.
      What it is forReading the economy as a whole: growth, inflation, unemployment, interest rates and debt, and judging what policy can and cannot do.
      Comes afterMicroeconomics (34/42).
      Leads toInternational Trade & Finance (34/47) and Development Economics (34/48), and it is the backdrop to the Business Administration major.
      What this piece gives youYou won't be able to forecast the next recession; nobody can reliably. You will understand why growth compounds, how inflation is measured, how banks make money, what a central bank does with its one lever, and why serious economists disagree about the rest. Easier first reads: Compound Interest Will Blow Your Mind (03/03), Where Public Money Goes (18/08) and The Mixed Economy (31/05).
    
  

  GDP, and what it misses
  Gross domestic product is the market value of everything produced inside a country in a year. You can count it three ways that must agree: add up what is produced, what is earned, or what is spent. The spending version is the one students memorise:
  GDP = C + I + G + (X − M)household consumption + business investment + government purchases + exports minus imports
  GDP is a superb tool for what it was built for: tracking output and jobs through a slump. Simon Kuznets, who built the first US national accounts, wrote in his 1934 report that "the welfare of a nation can scarcely be inferred from a measure of national income". It leaves out unpaid work (cooking, care for your parents), most environmental damage, and how income is shared. It also counts a rebuilt city after a flood as growth. Use it as a gauge of activity, not of well-being. Two standard adjustments matter: real GDP strips out price changes, and per capita divides by people, because a big country is not a rich one.

  Growth: compounding and the Solow idea
  Modest differences in growth rates, held for decades, make different worlds. Output per person growing at 1.6% a year doubles in 44 years; at 8.1% it doubles in about 9. The rule of thumb is to divide 70 by the percent. The table uses GDP per person, adjusted for price differences between countries, from the World Bank in constant 2021 international dollars.
  
    Country19902024Times largerPer year
    China1,66723,84114.38.1%
    India2,1409,4164.44.5%
    South Korea14,37855,0713.84.0%
    United States44,38475,6981.71.6%
    Nigeria5,4247,9941.51.2%
    Japan36,13847,4801.30.8%
    Mexico17,17021,9701.30.7%
  
  
    Interactive · The power of a few percent
    
      China, 1990-2024
      South Korea
      United States
      Mexico
    
    Growth per year
      
    Years
      
    
    
  
  Where does growth come from? The workhorse is the Solow model (Robert Solow, 1956; Nobel Prize 1987). An economy has workers, machines (capital) and technology. Saving builds new machines; machines wear out. But each extra machine adds less than the last, so with a fixed saving rate the economy settles at a level where new investment just replaces wear. Past that point more saving raises the level of income, not its growth. Lasting growth in income per person needs better technology and skills. Poor countries starting with little capital can grow fast by catching up, which is part of why China and India grew faster than the US.
  
    Interactive · Save more, or invent more?
    Share of income saved and invested
      
    Technology progress per year
      
    
    
    Toy model: output = technology × capital1/3, 5% of machines wear out each year, starting with little capital. Invented numbers.
  
  With 20% saving and 1% technology growth, output per worker climbs fast at first (year 5 growth is about 3.6%) and then settles to the pace of technology: about 1.1% by year 70, heading for 1%. Raise saving to 40% and the level ends higher, but the long-run growth rate is the same 1%. That is the Solow lesson, and why economists spend so much time on ideas, institutions and schooling. It is a model of mechanics, not a complete theory; Development Economics (34/48) is the argument about what else matters.

  Inflation: a basket, not a feeling
  Inflation is a sustained rise in the general level of prices. Statistical agencies measure it by pricing a fixed basket of goods a typical household buys (rent, food, fuel, transport) every month and tracking its cost. The weight of each item is its share of spending. Two consequences stay with you: your inflation differs from the headline whenever your basket differs, and the measure shifts when people switch goods.
  Real data, from the World Bank (average annual consumer prices, per cent):
  
    Country20222024
    Argentina72.4219.9
    Turkey72.358.5
    Nigeria18.833.2
    Brazil9.34.4
    United States8.02.9
    Germany6.92.3
    India6.75.0
    Japan2.52.7
  
  
    Interactive · Whose inflation?
    
      Average household
      Tight-budget renter
      Retired homeowner
    
    Energy price rise this year
      
    
    
    Invented baskets and price changes: housing +4%, food +9%, transport +7%, health & education +3%, other +2%, energy as set. Real baskets differ by country.
  
  With energy up 30% the average household's costs rise about 7.2%; the tight-budget renter's rise 8.4% and the retiree's 6.5%, because food and energy take a larger share. Most central banks, from the US to the euro area, aim for about 2% a year, a target chosen as a compromise between the harm of prices rising fast and the risk of falling prices, which make debts heavier and cause people to delay purchases.

  Unemployment: a flow, not a stock
  You are unemployed in the official sense if you have no job, are available, and have actively looked recently. People who have given up looking or can't find enough hours don't count, which is why statisticians also publish broader measures. The headline rate is the share of the labour force (workers plus job-seekers) out of work.
  What a graduate remembers is that it is a pool with a tap and a drain. Each month some employed people lose jobs and some unemployed people find them. If a fraction s of workers lose their jobs monthly and a fraction f of the unemployed find one, the pool settles at a rate of s ÷ (s + f). A rich-country economy can have lots of churn and a low rate, or little churn and a high one.
  
    PlaceRateYear
    Greece27.7%2013
    Spain26.1% (11.4% in 2024)2013
    South Africa32.3%2024
    United States14.8% (April 2020), 3.4% (April 2023)monthly
    Germany3.4%2024
    Japan2.5%2024
  
  
    Interactive · The unemployment pool
    Share of workers who lose a job each month
      
    Share of the unemployed who find one each month
      
    
    
    Steady-state rate = s ÷ (s + f), the textbook flow model. The numbers on the sliders are illustrative, not measured for any country.
  
  At 1.2% losing and 25% finding, the pool settles at 4.6%. Slow hiring by itself does the damage in a downturn: at 10% finding, the same layoffs give 10.7%. And the average spell is 1 ÷ f months, so a deep slump turns a short hunt into a long one. Economists distinguish frictional (between jobs), structural (skills or location mismatch) and cyclical (too little demand) unemployment, and the policy for each differs.

  Money and banks: where deposits come from
  Most of the money you hold is a number in a bank account. The old textbook story is the money multiplier: you deposit 100, the bank keeps a fraction as reserves and lends the rest, that loan becomes someone else's deposit, and so on. With a reserve ratio r the deposits can grow until they total the first deposit divided by r.
  
    Interactive · The textbook money multiplier
    
      5%
      10%
      20%
    
    Fraction kept in reserve
      
    
    
  
  At 10% a first deposit of 100 supports up to 1,000 of deposits; after 10 rounds you are at about 651. But the Bank of England's 2014 article "Money creation in the modern economy" (McLeay, Radia and Thomas) argues that the textbook story runs backwards: "the majority of money in the modern economy is created by commercial banks making loans", and banks "do not act simply as intermediaries, lending out deposits that savers place with them". A loan creates the matching deposit. What limits lending is then not reserves but borrowers' demand, banks' capital and the central bank's interest rate. That fits the fact that the US Federal Reserve set reserve requirements to zero in March 2020 with no sign of a lending explosion. The multiplier remains a good way to see that banks multiply money; it is a poor description of what restrains them.

  Central banks and the interest rate
  A central bank (the Federal Reserve, the European Central Bank, the Bank of Japan, Banco de México) sets the short-term interest rate and, through it, the price of borrowing for the whole economy. Raise it and mortgages, car loans and business investment get dearer; spending cools, and so does inflation. Cut it and the reverse. The effects arrive slowly, with a lag of a year or two, which is why steering resembles driving with a delayed windscreen.
  A simple benchmark is the Taylor rule (John Taylor, 1993): set the rate at inflation plus a neutral real rate of 2, nudged up by half the gap between inflation and the 2% target, and by half the output gap (how far the economy is above or below its sustainable level).
  rate = inflation + 2 + 0.5 × (inflation − 2) + 0.5 × output gapall in percent; a rule of thumb, not what any bank runs by formula
  
    Interactive · What would the rule say?
    
      Calm
      Overheating
      Slump
      9% inflation
    
    Inflation
      
    Output gap
      
    
    
  
  Calm gives 4%. A slump with 0.5% inflation and a gap of −6 gives −1.25%, below zero: the rule asks for a rate the bank can barely deliver, and in reality the ECB's deposit rate went only to −0.50% in 2019. (Negative rates are the edge of what central banks do; this is the "zero lower bound" and the reason unconventional tools exist.) The real data show the rule is a yardstick, not a law. In June 2022 US inflation was about 9% and the Fed funds rate 1.21%: it then rose to 5.33% by August 2023, and the ECB's deposit rate went from 0% in July 2022 to 4% in September 2023, while euro-area inflation peaked at 10.6% in October 2022. As of September 2026 the Fed's rate stood at 3.75% against US inflation of about 4.0%, and the ECB's at 2.5% against euro-area inflation of 3.75% (FRED).

  Fiscal policy and debt
  Where the central bank has one lever, governments have two: spending and taxes. In a slump they can spend more or tax less to prop up demand, and the effect on output can exceed the cost because the first dollar is someone's income, who spends part of it, and so on. The size of this multiplier is among the most argued numbers in the field. It depends on whether the economy has slack, whether the central bank offsets it, and how the spending is financed.
  The debt that results is a stock, and what matters is its size against income. Each year the ratio changes by the interest bill and growth: with r the interest rate and g the growth rate of the economy, debt as a share of GDP rises by roughly (r − g) times itself, less any primary surplus (revenue minus spending before interest).
  next debt ratio = debt ratio × (1 + r) ÷ (1 + g) − primary surplusall as shares of GDP; if growth exceeds the interest rate, debt shrinks even with a deficit
  
    Interactive · A country's debt over 30 years
    
      Growth keeps pace
      Interest outruns growth
      Austerity
    
    Starting debt (% of GDP)
      
    Interest rate on debt
      
    Growth (nominal GDP)
      
    Primary surplus (% of GDP)
      
    
    
    Toy model with constant rates. Real debt has maturities, currencies and changing interest rates.
  
  Starting at 100% with r = g, debt stays at 100%. If interest is 5% and growth 1%, debt climbs to 321% in 30 years; a primary surplus of 3% of GDP holds it to 154%. Real gross government debt (IMF, 2024): Japan 214% of GDP, Italy 135%, United States 122%, China 91%, South Africa 76%, Germany 61%, Mexico 59%. Japan shows that high debt in your own currency, mostly owed to your own citizens and with low interest, can last for decades. It does not show it is free: the bill is a constant claim on tax revenue.

  Business cycles, and the schools of thought
  Economies do not grow smoothly. They alternate expansions and recessions, and those swings are the business cycle: the US unemployment rate went from 3.5% in early 2020 to 14.8% in April 2020 and back to 3.4% by April 2023, an unusually sharp, fast cycle caused by a pandemic. The deeper question is why demand slumps persist and what to do. Four schools give different answers, and the field has not declared a winner.
  
    SchoolCore claimPolicy it favours
    Keynesian (Keynes, 1936; today's New Keynesians)Wages and prices are sticky, so demand shortfalls can leave people jobless for years.Governments and central banks should lean against slumps and booms.
    Monetarist (Friedman and Schwartz, 1963)"Inflation is always and everywhere a monetary phenomenon": money growth drives prices, and the Great Depression was worsened by the Fed letting the money supply shrink.Steady, rule-based money growth; central banks first fight inflation.
    New classical (Lucas, Nobel 1995)People form rational expectations, so they anticipate policy; the "Lucas critique" (1976) says models built from past data break when policy changes.Predictable rules over discretion; structural supply-side reform.
    Modern Monetary Theory (Wray, Mitchell, Kelton)A government that issues its own currency can't run out of it; the real limit is inflation and idle resources, not debt.Spend to full employment, tax to control inflation; job guarantee.
  
  Each has a record. Keynes explained what the 1930s needed; the 1970s stagflation, with high inflation and high unemployment together, embarrassed the simple Keynesian model. Monetarism helped end that inflation (Paul Volcker's money squeeze, 1979 to 1982) but the link between money and spending became unstable, so central banks moved to targeting inflation and setting interest rates directly. New classical ideas reshaped how models are built. MMT is a minority view: a 2019 University of Chicago IGM survey of leading economists unanimously rejected claims it attributed to MMT, while its supporters say they were being asked trick questions. The mainstream since the 1990s, "new Keynesian", blends sticky prices with rational expectations. Read the claims, not the labels.

  What a graduate just sees
  
    A few percent a year, compounded, separates rich countries from poor ones; divide 70 by the growth rate to get the doubling time.
    GDP counts activity, not welfare. Look at it per person and in real terms.
    Saving raises the level of income; only new ideas and skills raise its long-run growth.
    Your inflation is your basket's inflation, and a 2% target is a choice, not a fact.
    Unemployment is a pool: layoffs flow in, hiring flows out, and slow hiring is what makes slumps hurt.
    Banks create deposits when they lend. The brake is borrowers, capital and the policy rate, not a stock of reserves.
    A central bank has one lever, the short rate, and it works with a lag.
    Debt ratios move with r − g. Growth is a debt policy.
    Know which school is speaking, and what evidence would change its mind.
  

  Where it shows up next
  
    International Trade & Finance (34/47): the same accounts with an exchange rate; why currencies move and who wins and loses from trade.
    Development Economics (34/48): the growth table above, and the argument over why some countries stayed on the slow row.
    Public Economics (34/49): the other half of fiscal policy, what taxes and transfers do to people rather than to the aggregate.
    In the world: your mortgage rate, your wage negotiation, a company's investment decision, and the central bank announcement you read about this week.
  

  If you want to go deeper
  
    📘 Textbook. N. Gregory Mankiw, Macroeconomics (Worth/Macmillan). The standard intermediate text, with the Solow model and the schools set out clearly.
    📘 Textbook. Olivier Blanchard, Macroeconomics (Pearson). The other common choice, built around a single coherent model and recent episodes.
    📘 Gentler. Charles Wheelan, Naked Economics (W. W. Norton), or CORE Econ's free text The Economy, whose macro units follow growth, unemployment and inflation with real data.
    🌐 Free course. MIT OpenCourseWare 14.02 Principles of Macroeconomics (Spring 2023). Also Khan Academy Macroeconomics.
  

  
    The takeaway
    An economy is a few connected pipes: output that compounds, prices that drift, jobs that flow in and out, money that banks create by lending, and debt that moves with the gap between interest and growth. Central banks have one lever and governments two; none of them can promise a smooth ride. The schools differ on how much stabilising is wise, and the honest answer is to ask what each would predict.
    Try this: this week, find your own inflation rate. List the five things you spend most on, look up how each price changed in the last year, and compare with the headline figure for your country.
  

  
    Sources & Further Reading
    Sources. Concepts: standard textbook macroeconomics (Mankiw; Blanchard). Data (accessed November 2026): World Bank WDI (NY.GDP.PCAP.PP.KD, constant 2021 international dollars, 1990 and 2024; FP.CPI.TOTL.ZG; SL.UEM.TOTL.ZS, ILO modelled); FRED for US CPI (June 2022 about 9.0% on the seasonally adjusted series, a tenth below the BLS headline), unemployment (14.8% April 2020, 3.4% April 2023), the Fed funds rate (1.21% June 2022, 5.33% August 2023, 3.75% September 2026), the ECB deposit rate (−0.50% in 2019, 0% July 2022, 4.00% September 2023, 2.50% September 2026) and euro-area HICP (Eurostat: peak 10.6% October 2022, 3.75% September 2026); IMF DataMapper, general government gross debt (WEO). Bank of England, McLeay, Radia and Thomas, Quarterly Bulletin 2014 Q1; Federal Reserve Board, reserve requirements 0% from 26 March 2020; Kuznets (1934); Solow (1956, Nobel 1987); Taylor (1993); Lucas (1976, Nobel 1995); Friedman and Schwartz (1963); Chicago Booth IGM survey (2019) on MMT. Computed in node: 1.6% for 34 years is 1.72x, 8.1% is 14.1x; basket results 7.2%, 8.4%, 6.5%; unemployment 4.6% and 10.7%; Taylor 4.0% and −1.25%; debt 321% and 154%. Baskets, weights and shocks are invented.
    
      🌐 MIT OCW 14.02 Principles of Macroeconomics
      🌐 Bank of England: Money creation in the modern economy
      🌐 CORE Econ, The Economy
      📘 Mankiw, Macroeconomics


Step 4. Probability & Statistics: Reasoning with Uncertainty
------------------------------------------------------------
Why: Economic data is noisy; statistics is how economists avoid fooling themselves.
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 5. Linear Algebra: Machines That Move Space
------------------------------------------------
Why: Many markets at once, and every regression, are linear algebra.
A grid of numbers is really a machine that moves all of space at once. Learn to see it, and circuits, bridges, search engines and AI all start to look alike.

You pinch two fingers on a photo and it rotates and zooms smoothly under them. A game character turns to face you and every corner of its face swings round together. A search for "best pizza near me" ranks millions of pages in a fraction of a second. A bridge designer presses "solve" and a screen of colour shows how every beam is being squeezed.
  Underneath all four is the same object: a grid of numbers, and one habit of thought about what a grid of numbers does. A graduate of a linear algebra course no longer sees a table of numbers. They see a machine that grabs the whole plane, or the whole of space, and moves it.
  This piece builds that picture with seven things you can drag. You will not come out able to compute a determinant of a 5×5 by hand (nobody does that on purpose). You will come out seeing why a system of equations sometimes has no answer, why a bridge hums at particular pitches, and why the chips that run AI are built for one operation only.

  
    The course in one box
    Who takes it: nearly every engineering and computer science student, usually in the first or second year. It is the single most reused maths course in both majors.
    What it's for: handling many unknowns and many dimensions at once, with one set of tools.
    Comes after: only school algebra and a little geometry. (Calculus helps but is not required.)
    Leads to: Statics (34/10), Control Systems (34/20) and Machine Learning (34/32) list it as a prerequisite; Vibrations (34/17), Computer Architecture (34/25) and Multivariable Calculus (34/02) lean on it constantly.
    Honest line: you won't be able to run a finite element program after this piece. You will see why it spends nearly all its time solving one equation, Ax = b. If you want an easier first read, try the library's Vectors, Matrices and Linear Transformation (12/14); this one goes further into solving, eigenvectors, fitting and why hardware cares.
  

  A matrix moves all of space at once
  Start with the picture a graduate keeps. A 2×2 matrix is not four numbers. It is two arrows. The first column says where the arrow pointing one step right (call it i) ends up. The second column says where the arrow pointing one step up (j) ends up. Once you know where those two land, you know where everything lands, because every point is some amount of i plus some amount of j, and the machine is straight and even: grid lines stay straight, and evenly spaced lines stay evenly spaced.
  Ax = x1(column 1) + x2(column 2)
  Read that in words: the matrix times a point is "that many of the first column, plus that many of the second". Try it. The grey grid is the original plane; the blue grid is where the matrix sends it.

  
    Interactive · Move the plane
    
      Do nothing
      Rotate 30°
      Stretch
      Shear
      Mirror
    
    
      Column 1, right part (a)
      Column 2, right part (b)
      Column 1, up part (c)
      Column 2, up part (d)
      
      
    
    Without JavaScript: the rotate-30° matrix sends the arrow (1, 0) to (0.87, 0.5) and (0, 1) to (−0.5, 0.87). Every grid line turns 30° and the unit square keeps area 1.
  
  Notice what the shaded unit square does. In "Rotate" it keeps its size. In "Stretch" it doubles. In "Shear" it leans over but keeps its area. That area change is the next idea.

  The determinant is how much area gets scaled
  The determinant is one number that says how much the machine scales area. If the unit square becomes a shape with area 5, the determinant is 5. If it becomes a mirror image, the determinant is negative: area is kept (or scaled) but orientation is flipped. For a 2×2 matrix with columns (a, c) and (b, d) the rule is short:
  det = ad − bc
  Worked number: the matrix with columns (2, 1) and (1, 3) has det = 2×3 − 1×1 = 5, so it stretches every area in the plane by a factor of five. In three dimensions the determinant scales volume.
  The interesting case is zero. A determinant of zero means the machine flattens space: the plane is squashed onto a line (or a point), and the area becomes nothing. Squashing cannot be undone, because many different starting points land on the same spot. Drag the slider until the determinant reaches zero.

  
    Interactive · Squash the plane flat
    
      Tilt the second arrow: s
      
      
    
    Without JavaScript: the matrix with columns (1, 0.5) and (s, 1) has determinant 1 − 0.5s. At s = 2 the determinant is 0, the whole plane lands on one line, and the points (0, 0) and (2, −1) both land on the origin.
  
  The orange dot is a probe at (2, −1). At s = 2 it lands exactly on the origin, the same place as the point (0, 0). Two different inputs, one output: no machine could tell you which one you started from. That is the whole meaning of "this matrix has no inverse", and it is why a zero determinant is the alarm bell in every solver you will ever meet.

  Solving many equations at once: Ax = b
  Now use the machine backwards. Given the output b, which input x did the machine start from? That is the equation Ax = b, and it is the most common problem in applied mathematics. Every time you have several unknowns that are tied together by several straight-line relationships, you have one.
  A circuit is the cleanest example. Two loops share a resistor. Kirchhoff's voltage law says the voltage gained around each loop equals the voltage lost, which gives one equation per loop. With a 10 V battery, R1 = 2 Ω, R2 = 3 Ω (shared) and R3 = 4 Ω:
  5I1 − 3I2 = 10−3I1 + 7I2 = 0
  The determinant is 5×7 − 3×3 = 26, so there is exactly one answer: I1 = 2.69 A and I2 = 1.15 A. Change the resistors below and watch the matrix and the answer change together.

  
    Interactive · Solve a circuit with a matrix
    
      Battery V
      R1 (loop 1 only)
      R2 (shared)
      R3 (loop 2 only)
      
      
    
    Without JavaScript: with 10 V, 2 Ω, 3 Ω and 4 Ω the equations are 5I1 − 3I2 = 10 and −3I1 + 7I2 = 0, its determinant is 26, and the loop currents are 2.69 A and 1.15 A.
  
  The reason an engineer cares: nothing about this method knows it is a circuit. A truss is the same. At each joint of a bridge, the horizontal forces must add to zero and so must the vertical ones: two equations per joint. A simple triangle truss has three joints and so six equations, for three member forces and three support reactions, and the answer is one solve of Ax = b. A real structure has thousands of joints. Finite element programs routinely solve systems of many thousands of equations, and research codes have reported solving over a million unknowns at once. The method is the same one you just used; only the size has changed. (Statics, 34/10, is this idea applied to forces.)
  A zero determinant in a structure's equations is a warning: the structure may be able to move without any member pushing back.

  Eigenvectors: the directions a transformation only stretches
  Most arrows are turned by a matrix as well as stretched. But for many matrices there are special directions where the arrow comes out pointing the same way (or exactly backwards), only longer or shorter. Those are the eigenvectors, and the stretch factor along each is its eigenvalue. (Eigen is German for "own": the matrix's own directions.)
  Av = λv
  In words: feed the matrix the special arrow v and it hands back the same arrow multiplied by a plain number λ. For the matrix with columns (2, 1) and (1, 2) the special directions are 45° (stretched by 3) and 135° (stretched by 1). Every other arrow gets bent towards the stretchy one. Turn the dial and look for the moments the two arrows line up.

  
    Interactive · Find the directions that only stretch
    
      Stretch & mix
      Shear
      Rotation
      Two springs
    
    
      Direction of the input arrow
      
      
    
    Without JavaScript: for the matrix with columns (2, 1) and (1, 2), an arrow at 45° comes out at 45° three times longer, and an arrow at 135° comes out at 135° unchanged. A pure rotation has no such direction at all.
  
  Three things to notice. The shear has only one special direction (along the x-axis, where nothing happens at all). The rotation has none, because every arrow turns. And "Two springs" has two special directions with different stretch factors, 1 and 3; hold that thought.

  Eigenvectors at work: bridges that hum, pages that rank
  Vibration. Picture two equal masses on springs between two walls. Push them any way you like and the motion looks messy. But there are exactly two clean patterns: both swing together, or they swing in opposite directions. These are the modes, and they are the eigenvectors of the "two springs" matrix above, (1, 1) and (1, −1). The eigenvalues 1 and 3 fix how fast each mode vibrates: the squared frequency is the eigenvalue times k/m, so the opposite-motion mode vibrates √3 ≈ 1.73 times faster. Any messy motion is just a mix of the modes. That is how engineers find which pitches will shake a wing, a bridge deck or a drive shaft (Vibrations, 34/17).
  Ranking pages. Google's founders modelled a web surfer who follows a random link from each page, and now and then jumps to a random page instead. The long-run share of time spent on each page is its PageRank, and it is the eigenvector (with eigenvalue 1) of the matrix describing one step of that surfer. In their 1998 paper Brin and Page used a "damping" probability of 0.85 for following links. You can find the eigenvector without solving anything fancy: start with equal scores and apply the machine over and over. Press Step.

  
    Interactive · A four-page web finds its eigenvector
    
      Step onceRun 20 stepsReset
      
      
    
    Without JavaScript: A links to B and C, B links to C, C links to A, D links to C. After many steps with damping 0.85 the ranks settle to A 0.373, B 0.196, C 0.394, D 0.038. D, which nobody links to, keeps only the base share.
  
  After about twenty steps the numbers stop changing: the vector has become an eigenvector, because applying the machine again returns the same vector. Page C wins because two pages point at it and one of them (A) is itself well-ranked.

  Least squares: the best answer when there is no exact one
  Real data is never perfect. Six measurements that "should" lie on a straight line will not, so the equations Ax = b have more equations than unknowns and no exact solution. Do you give up? No. You choose the line whose total mistake is smallest. Measure each miss vertically, square it, and add up the squares. The least squares line makes that total as small as possible.
  In pictures: each squared miss is literally a square. Drag the line and watch the squares shrink and grow.

  
    Interactive · Make the squares as small as possible
    
      Slope
      Height at x = 0
      Snap to the least-squares line
      
      
    
    Without JavaScript: for the six points (1, 1.2), (2, 2.9), (3, 3.1), (4, 4.8), (5, 5.2), (6, 7.1) the best line is y = 1.089x + 0.24, with total squared miss 0.798. The guess y = x gives 2.75.
  
  The graduate's picture is geometric: the right-hand side b is a point outside the flat sheet of everything the matrix can reach, and the best answer is the point on the sheet closest to it, found by dropping a perpendicular. That perpendicular condition gives the one equation to remember:
  ATA x̂ = ATb
  In words: multiply both sides by the matrix flipped on its side and you get a square system you can solve. The history is a priority fight: Legendre published the method in 1805, Gauss said he had used it since 1795 and published in 1809, and Gauss's use of it to predict where the asteroid Ceres would reappear in 1801 made it famous. Today it sits under every trend line, every calibration of a sensor, and the first model most machine-learning students ever train.

  Why GPUs and AI are matrix multiplication
  Multiplying two matrices is just many dot products: every cell of the answer is one row of the first matrix paired with one column of the second, multiplied and added. Click a cell below and see the row and column that make it.

  
    Interactive · Build a product cell by cell
    
      Tap a cell of C = A × B.
      
      
      Matrix size n (n × n times n × n)
      
        1 billion multiplies per second
        100 billion per second
        10 trillion per second
      
      
      The speeds are round illustrative numbers, not any particular chip.
    
    Without JavaScript: multiplying two n×n matrices takes n×n×n multiplications. For n = 1,000 that is 1,000,000,000, and the 1,000,000 cells of the answer can all be computed independently.
  
  Two facts follow. First, the work is huge: n×n matrices need n³ multiplications, so 1,000×1,000 is a billion. Second, the cells are independent, so a thousand workers (or a million) can each take cells at the same time. That is exactly the shape of work a graphics chip was built for, because drawing a 3D scene means pushing every corner of every triangle through a matrix that rotates, moves and projects it.
  Modern neural networks turned out to be the same shape. A layer of a network is a matrix of learned numbers times a vector (or a matrix) of inputs. In their scaling-law paper, Kaplan and colleagues estimate that a transformer language model with N parameters needs roughly 2N arithmetic operations to process one token. GPT-3, with 175 billion parameters, therefore needs about 3.5×1011 operations for each token, nearly all of them inside matrix products. NVIDIA's Volta GPUs (2017) put dedicated "tensor core" circuits on the chip that do nothing but a small matrix multiply-and-add (4×4) per clock tick. When people say AI runs on GPUs, they mean it runs on hardware shaped like linear algebra. (Computer Architecture, 34/25, and Machine Learning, 34/32.)

  What a graduate just sees
  
    A matrix is a machine that moves all of space; its columns say where the basis arrows go.
    The determinant is the area (or volume) scale factor, and zero means the machine flattens space and cannot be undone.
    "Solve Ax = b" is the same question for a circuit, a truss, a fitted line and a game physics step.
    If there are as many independent equations as unknowns there is one answer; fewer, many answers; contradictory, none.
    Eigenvectors are the directions a machine only stretches; they simplify any problem the machine drives.
    A vibrating structure is a mix of modes, one per eigenvector; the eigenvalues are the pitches.
    When there is no exact answer, take the closest one: project onto what you can reach.
    Matrix multiplication is many independent dot products, which is why it parallelises and why hardware is built around it.
  

  Where it shows up next
  
    Statics (34/10) and Mechanics of Materials (34/11): joint and member forces are a linear system; stiffness matrices tie loads to deflections.
    Vibrations (34/17) and Dynamics (34/12): mode shapes and natural frequencies are eigenvectors and eigenvalues of mass and stiffness matrices.
    Control Systems (34/20): the eigenvalues of a system's matrix say whether a drone's wobble dies away or grows.
    Machine Learning (34/32): data is a matrix, a layer is a matrix, and training adjusts matrices; least squares is the model you meet first.
    Computer Architecture (34/25) and graphics: GPUs and tensor cores exist to multiply matrices fast.
  

  If you want to go deeper
  
    📘 Gilbert Strang, Introduction to Linear Algebra (Wellesley-Cambridge Press; 6th edition, 2023). The standard first course for engineers; best for the four fundamental subspaces and for application.
    📘 David C. Lay, Steven R. Lay and Judi J. McDonald, Linear Algebra and Its Applications (Pearson; 6th edition, 2021). Gentle, example-heavy, widely assigned. For proofs, Sheldon Axler's Linear Algebra Done Right (Springer; 4th edition, free to read online at linear.axler.net) is the elegant theoretical route.
    ▶️ 3Blue1Brown, Essence of Linear Algebra: the best visual series for learning alone; it builds exactly the "matrix as moving space" picture used here.
    🌐 MIT OpenCourseWare 18.06, Linear Algebra (Gilbert Strang): full video lectures, free.
  

  
    The takeaway
    Linear algebra is the study of machines that move all of space at once. Read a matrix by its columns, read its determinant as the area scale, solve Ax = b when you need to run the machine backwards, look for the eigenvector directions when you want it to be simple, and project onto what you can reach when there is no exact answer.
    Try this: next time an app zooms, rotates or recolours a photo, picture the two column arrows moving. And for the fun of it, in the circuit widget raise R2 to 10: the shared resistor starts to decide what both loops do.
  

  
    Sources & Further Reading
    Sources. Strang, Introduction to Linear Algebra (6th ed., 2023) and MIT 18.06 for the standard course content. Kirchhoff loop equations and the spring-mass modes are textbook derivations; every number above was recomputed in code. Legendre's 1805 publication and Gauss's priority claim and Ceres prediction: Wikipedia, "Least squares"; MacTutor biography of Legendre. PageRank and the 0.85 damping factor: Brin and Page, "The Anatomy of a Large-Scale Hypertextual Web Search Engine" (1998). Volta tensor cores (4×4 matrix multiply-accumulate): Markidis et al., "NVIDIA Tensor Core Programmability, Performance & Precision" (2018). About 2N operations per token: Kaplan et al., "Scaling Laws for Neural Language Models" (2020); GPT-3 at 175 billion parameters: Brown et al. (2020). Million-unknown structural solves: NASA Technical Reports Server and OSTI conference papers on parallel sparse solvers.
    Go deeper. The books and courses above.


Step 6. Multivariable Calculus: Calculus on a Landscape
-------------------------------------------------------
Why: Choosing the best bundle under a budget is optimisation in many variables.
A raindrop on a hillside picks the steepest way down without thinking. The arrow it follows is the same one that trains every neural network and drives heat through a metal plate.

Stand on a hillside with a map. Height depends on two things at once: how far east you are and how far north. Single-variable calculus handles one input at a time. The hillside needs a new question: of all the directions I could step, which one rises fastest?
  Almost nothing real depends on one number. Temperature in an engine block depends on position in three dimensions. The error of a language model depends on billions of adjustable weights. Air pressure depends on place and time. This course takes the slope-and-area ideas of Calculus I (34/01) and lets them live in more than one dimension. Surprisingly little new is needed: one new arrow, the gradient, and one new accounting rule, the idea that what flows out of a region is what is made inside it.

  
    The course in one box
    Who takes it: second-year Mechanical & Aerospace Engineering students; Computer Science students meet it in year 4 when machine learning needs it.
    What it is for: calculating change and totals when many things vary at once.
    Comes after: Calculus I (34/01).
    Leads to: Physics II: Electricity & Magnetism (34/07), Fluid Mechanics (34/15), Machine Learning (34/32), and through them Heat Transfer (34/16) and Propulsion (34/21).
    Honest line: you will not be able to grind through a surface integral after this. You will see why a gradient points uphill, why training an AI is "just" rolling downhill, and what the big theorems say in plain words.
  

  Two inputs make a landscape, and a partial derivative is one slope
  A function of two inputs is a surface: height above each point of a map. Engineers draw it the way hikers do, with contour lines: closely spaced lines mean steep ground, widely spaced lines mean gentle ground.
  To ask "how steep?" you must say in which direction. Walk due east and note the slope: that is the partial derivative with respect to x, written ∂f/∂x. Hold east still and walk north: that is ∂f/∂y. The recipe is simple: treat every other variable as frozen and differentiate as before.
  ∂f/∂x = slope of f as only x changes, with y held still
  The landscape below has a tall hill near (1, 1) and a lower one near (−1.2, −0.8). Drag the point on the map (or use the sliders). The cyan arrow shows the slope eastward and the green arrow the slope northward; a leftward or downward arrow means the ground falls in that direction.

  
    Interactive · Walk the map
    
    East–west position x
      
    North–south position y
      
    
  
  At the starting point (−0.2, 0.3) the height is 0.684, the eastward slope +0.621 and the northward slope +0.226: the ground rises both ways, more steeply towards the east.

  The gradient points straight uphill
  Put the two partial slopes together as one arrow, east component first: that is the gradient, written ∇f. It has a beautiful property: it points in the direction of steepest ascent, and its length is how steep that is. Step in any other direction and the slope is the gradient's length times the cosine of the angle between your step and the arrow. Straight along it, cosine is 1: the most you can get. Sideways to it, cosine is 0: you walk along a contour and stay level.
  slope in direction u = ∇f · u = |∇f| cos(angle)
  Two consequences follow. The gradient is always perpendicular to the contour line through that point, which is why water runs across contours, not along them. And the same arrow governs heat: in Fourier's law, heat flows down the temperature gradient, from hot to cold, at a rate proportional to its steepness. Many diffusion laws have this shape: flow follows the negative gradient.

  
    Interactive · Which way is uphill?
    
    Direction you step (0° = east, 90° = north)
      
    Turn to the gradient
    
  
  Turn the pink direction until the slope is largest: it lines up with the orange arrow. Turn it 90° from the arrow and the slope is zero. At the very top of a hill the gradient shrinks to nothing, which is the multivariable version of "zero slope means a peak" from Calculus I.

  Rolling downhill is how a network learns
  Now flip the sign. A machine-learning model has adjustable numbers (weights), and a loss measures how wrong it is on the training data. The loss is a landscape over the weights, with one dimension per weight. To learn, take a small step against the gradient, so downhill, then recompute and repeat. This is gradient descent, which Augustin-Louis Cauchy proposed in 1847 for solving equations, long before computers. The chain rule from Calculus I is what lets the gradient be computed layer by layer, in the method called backpropagation, published by Rumelhart, Hinton and Williams in Nature in 1986. A model like GPT-3 (2020) has 175 billion weights, so its landscape has 175 billion dimensions, yet the rule is the same arrow.
  new position = old position − (step size) × ∇(loss)
  The step size matters, and you can feel why. In the bowl below, one direction is twelve times steeper than the other. Too small a step crawls; just right is fast; too large and each step overshoots the valley floor by more than it started and the loss explodes. For this bowl, steps above 1/6 (about 0.167) diverge.

  
    Interactive · Learning rate
    
      Too small
      Good
      Zig-zag
      Too big
    
    Step size (learning rate)
      
    Steps taken
      
    
    
  
  Real training uses refinements (momentum, adaptive step sizes, random mini-batches), but all of them are ways of coping with exactly this picture: a long narrow valley where the steepest direction is not the direction of the destination.

  Double and triple integrals add up a landscape's worth
  Integrals extend the same way. A single integral adds thin slices along a line. A double integral adds tiny patches over an area: if each patch has a density (kilograms per square metre), the sum is the plate's mass. A triple integral does the same over a solid, with density in kilograms per cubic metre. The recipe is slice, multiply, add, repeated once per dimension.
  mass = sum over all pieces of (density × piece size)  =  ∬ ρ dA
  The same sums also give the centroid, the balance point: weight each position by its piece's mass and divide by the total. Engineers use it constantly: the centre of mass of a rocket, a car or an aircraft decides whether it is stable. Try a plate 2 m long and 1 m wide whose density rises along the length.

  
    Interactive · Where does the plate balance?
    Density rises along the length: a (kg per m² per m)
      
    Density rises across the width: b
      
    
    
  
  The density here is 1 + ax + by kg/m². With a = 3 and b = 0, the plate's mass is 2 + 2×3 = 8 kg and it balances at x = 1.25 m, shifted from the middle (1 m) towards the heavy end.

  A vector field puts an arrow on every point
  Wind on a weather map, water in a pipe, the pull of a magnet: each attaches an arrow (a direction and strength) to every point in space. That is a vector field. Two numbers computed from the field's partial derivatives tell you most of what it is doing locally.
  Divergence: is the field spreading out from here, or converging? A tap under water is a source; a drain is a sink. A positive divergence means a small blob of fluid is expanding. Curl: is the field spinning? Put a tiny paddle wheel in the flow. If it turns, the curl is not zero, and the sign tells you which way.
  divergence = ∂vx/∂x + ∂vy/∂y    curl = ∂vy/∂x − ∂vx/∂y
  Pick a field and let a round blob of fluid (grey dashes show where it started) be carried along. You will see that a source inflates the blob without turning it, a whirlpool turns it without changing its area, and a saddle stretches it one way and squeezes it the other while keeping its area.

  
    Interactive · Drift a blob through a field
    
      Source
      Whirlpool
      Shear
      Saddle
    
    Time elapsed
      
    
    
  
  Divergence and curl are the two words in Maxwell's equations of electromagnetism (Physics II, 34/07) and in the equations of fluid flow (Fluid Mechanics, 34/15). Spreading and spinning are the two things a field can do locally, besides moving along as a whole.

  What flows out of a region is what is made inside
  Here is the idea behind the three big theorems, in the language of a bathtub. If water pours out through the boundary of a region, then some taps inside must be producing it (or the level inside must be falling). Add up every tiny source and sink inside and you get the net flow out through the boundary. Nothing more is needed.
  net flow out through the boundary = total divergence inside
  That is the divergence theorem, the three-dimensional version of the statement. Green's theorem is the flat version (the name honours George Green, whose Essay of 1828 contained the related surface-to-volume formula), and Stokes's theorem relates the spinning of a field inside a surface to the flow around its edge. All three say the same kind of thing: a sum over a boundary equals a sum of local changes inside. It is why one can measure a rotating machine from outside and why conservation laws can be written either as "this region's total changes by what crosses its walls" (integral form) or as an equation about each point (differential form).

  
    Interactive · Count the flow two ways
    
      Source
      Cubic
      Source in, sink out
      Steady wind
    
    Circle radius
      
    Circle centre, left or right
      
    
    
  
  The first number adds up the field's push through the circle itself. The second never looks at the circle's edge: it adds the divergence in every patch inside. They agree for every field, radius and position. For the plain source (x, y), the divergence is 2 everywhere, so a unit circle encloses 2 × π = 6.283 units of production, and exactly 6.283 flows out through its rim.

  What a graduate just sees
  
    A slope needs a direction; a partial derivative is the slope with every other input held still.
    The gradient points straight uphill, its length is the steepness, and it is perpendicular to the contour.
    "Downhill" is the whole idea behind training a model: step against the gradient, and mind the step size.
    Heat, and many other things, flow down a gradient from high to low.
    A double or triple integral is slice, multiply, add, once per dimension; mass and balance point come from it.
    A field can only spread (divergence) or spin (curl), locally.
    What leaves a region through its wall is exactly what its insides produce.
  

  Where it shows up next
  
    Physics II: Electricity & Magnetism (34/07): Maxwell's equations are statements about the divergence and curl of the electric and magnetic fields.
    Fluid Mechanics (34/15): conservation of mass is a divergence equation, and the divergence theorem turns "what crosses the walls of this control volume" into "what happens inside it".
    Heat Transfer (34/16): conduction is heat flowing down the temperature gradient; the heat equation is the gradient and divergence together.
    Machine Learning (34/32): training is gradient descent on a loss landscape with millions or billions of dimensions.
    Mechanics of Materials (34/11) and Dynamics (34/12): centroids and moments of inertia are integrals of exactly the kind you added up above.
  

  If you want to go deeper
  
    📘 Textbook: Stewart, Clegg and Watson, Calculus: Early Transcendentals (9th ed.), chapters 14 to 16: the standard multivariable text, rich in examples.
    📘 Textbook: Jerrold Marsden and Anthony Tromba, Vector Calculus (6th ed., W. H. Freeman): the classic for the geometry of divergence, curl and the big theorems.
    ▶️ Gentler: Khan Academy's Multivariable Calculus course, which has videos with a strong visual feel, and Grant Sanderson's multivariable lessons.
    🌐 Free course: MIT OpenCourseWare 18.02SC Multivariable Calculus, with lectures, notes and exercises.
  

  
    The takeaway
    With many inputs, a slope needs a direction. The partial derivatives are the slopes along the axes; the gradient joins them into one arrow that points uphill and measures how steep it is. Walk against it and you are doing gradient descent, which trains models with billions of weights and also describes how heat flows. Integrals extend to areas and volumes, giving mass and balance points, and for flows the whole subject rests on one bookkeeping rule: what leaves a region through its boundary is what its interior produces.
    Try this: on any hill, a park slope or a street, find the line along which a ball would roll: that is the negative gradient. Then notice that the ball's path crosses the contours at right angles.
  

  
    Sources & Further Reading
    Sources. Gradient descent: A.-L. Cauchy, "Méthode générale pour la résolution des systèmes d'équations simultanées", Comptes rendus 25 (1847), pp. 536–538; Claude Lemaréchal, "Cauchy and the gradient method", Documenta Mathematica, Extra Volume ISMP (2012). Backpropagation: Rumelhart, Hinton and Williams, Nature 323 (1986), pp. 533–536. GPT-3's 175 billion parameters: Brown et al., "Language Models are Few-Shot Learners" (2020, arXiv 2005.14165). Green's Essay of 1828 containing the theorem named after him: MacTutor biography and the essay itself. Computed in node: the landscape f = 2e−((x−1)²+(y−1)²)/1.5 + 1.2e−((x+1.2)²+(y+0.8)²)/1.0 and its analytic gradient (checked against finite differences, agreement to 3 × 10−10); gradient descent on ½(x² + 12y²), stable for step sizes below 2/12; plate mass and centroid by 3,200-cell sums against the exact formulas; blob area against ediv·t; circle flux against the divergence integral for four fields (agreement to 4 decimals).
    
      🌐 MIT OpenCourseWare 18.02SC Multivariable Calculus
      🌐 Khan Academy: Multivariable calculus
      🌐 Documenta Mathematica, ISMP volume, which contains Lemaréchal's short history of gradient descent
      📘 Stewart, Clegg and Watson, Calculus: Early Transcendentals, 9th ed. (Cengage)
      📘 Marsden and Tromba, Vector Calculus, 6th ed.


Step 7. Econometrics: How to Find What Caused What
--------------------------------------------------
Why: Correlation is cheap; the credibility revolution is how economists find causes.
Correlation is cheap. Econometrics is the craft of getting from "these move together" to "this made that happen", and of knowing how badly you could be fooled on the way.

A city opens a new tram line, and a year later house prices along it have risen. Did the tram do it? Maybe. Or maybe the city built the line where it already expected the area to boom. Or a new university opened nearby the same year. Every headline that says "X is linked to Y" hides this question, and most of the time the headline cannot answer it.
  An economist sees the same pattern in a minimum-wage law, a school-class size, a tax change, a drug trial. The data show a link. The job is to say what the link would have been if nothing else had been going on. That is econometrics: statistics built for questions where you usually cannot run the experiment, and where the people in your data chose their own circumstances.

  
    The course in one box
    
      Who takes itEconomics majors, usually in year 2 or 3, straight after a first statistics course. Public policy, political science, finance and business-analytics programmes teach close cousins. Programmes such as MIT's Course 14-1, the LSE and Chicago all require it.
      What it is forEstimating the effect of one thing on another from messy real-world data, with an honest statement of how wrong the estimate might be.
      Comes afterProbability & Statistics (34/05).
      Leads toDevelopment Economics (34/48), and it is the working tool in Behavioral Economics (34/46), Marketing Research & Analytics (34/59) and Machine Learning (34/32).
      What this piece gives youYou will not be able to run a regression package after this. You will see why a coefficient can lie, what a clever study does to stop it lying, and why a "significant" result is weaker evidence than it sounds. It goes further than Probability & Statistics (34/05), whose last section stops at "correlation is not causation"; here is what to do about it. The library's How to Spot Bullshit (17/03) is a gentler start. All datasets here are invented and labelled illustrative.
    
  

  Regression is the best-fitting line
  Take twelve students: hours studied, and the score each got. Plot them and the dots drift upward. A regression draws the one straight line that sits closest to all the dots. "Closest" has a precise meaning: for each dot, measure the vertical gap to the line (the residual), square it, and add up the squares. The line with the smallest total is the ordinary least squares (OLS) line. Squaring means one big miss counts far more than several small ones; the picture shows each squared miss as a literal square.
  score = a + b × hours + errorb is the slope: the score gap between two students who differ by one hour
  
    Interactive · Fit the line yourself (illustrative data)
    Slope b (points per hour)
      
    Intercept a (score at 0 hours)
      
    Snap to the best line
    
    
    Twelve invented students. The best line is found by solving the least-squares equations; no other line has a smaller sum of squared misses.
  
  The best line here is about 47 + 3.5 × hours (the readout gives it exactly). The coefficient b says: students who studied an hour more scored about 3.5 points higher. Notice what that sentence does not say. It compares different students. It does not say that making one student study an extra hour would raise her score by 3.5. That gap, between a comparison and a cause, is the rest of the course.

  Omitted variables: add the confounder, watch the effect change
  Suppose you regress wages on years of schooling in a big survey and find that each extra year goes with higher pay. Is that what schooling does? People with more drive or talent may both stay in school longer and earn more anyway. If so, the schooling coefficient is also soaking up the talent effect. This is omitted-variable bias, and it has a tidy rule:
  bias ≈ (effect of the omitted factor on the outcome) × (how strongly it moves with your variable)if either is zero, there is no bias
  In the lab below, the true effect of one more year of schooling is exactly 3 (thousand dollars). Talent also raises pay, and a dial sets how strongly talent pushes people toward more school. Switch the model between ignoring talent and including it.
  
    Interactive · The hidden confounder (illustrative simulation, 1,500 people)
    How strongly talent pushes people to stay in school
      
    
      Ignore talent
      Control for talent
      Randomise schooling
    
    
    above-average talent   below-average talent
    
    True schooling effect = 3.0; talent adds 8 per standard deviation. Wage = 10 + 3·school + 8·talent + noise. "Randomise" means a coin-flip-like process assigns schooling regardless of talent.
  
  Without JavaScript: with talent pushing schooling at strength 1.2, ignoring talent gives a schooling coefficient near 4.7 instead of the true 3.0; controlling for talent recovers about 3.0; randomising schooling also recovers about 3.0.
  Ignoring talent at strength 1.2 gives about 4.7, over half again too large. Control for talent and it falls to about 3. The cost of this fix: you must measure the confounder. Most real confounders (motivation, family connections, how well-run the firm is) sit in no dataset. That is why economists spent decades arguing about what regressions could prove.

  Randomise: the coin that cuts every hidden link
  Press the third button. When schooling is assigned by a process unrelated to talent, the link between them is broken, and even the naive regression that ignores talent lands on 3. A coin flip is independent of everything: motivation, family, luck, things nobody thought to measure. That is why a randomised experiment is the gold standard, and why 2019's Nobel prize in economics went to Abhijit Banerjee, Esther Duflo and Michael Kremer for experimental approaches to reducing poverty.
  But economists rarely get to flip the coin: you cannot randomly assign years of schooling, a recession or a minimum wage. The next three ideas are ways to find a coin that nature or a rule has already flipped. Together they are called the credibility revolution, a phrase from Joshua Angrist and Jörn-Steffen Pischke's 2010 essay, answering Edward Leamer's 1983 complaint that econometrics needed to "take the con out" of itself.

  Difference-in-differences: borrow the other group's trend
  In 1992 New Jersey raised its minimum wage and neighbouring Pennsylvania did not. David Card and Alan Krueger surveyed fast-food restaurants in both states before and after (American Economic Review, 1994). Looking only at New Jersey before and after would confuse the law with anything else changing that year. Instead they used Pennsylvania as the stand-in for "what would have happened anyway". Their finding, that employment did not fall in New Jersey, contradicted the textbook prediction at the time. It has been disputed ever since: Neumark and Wascher re-analysed it with payroll records and found job losses, and the field's overall reading is still argued over. The design is not in dispute. Card shared the 2021 Nobel prize in part for it.
  effect = (treated after − treated before) − (comparison after − comparison before)subtract the comparison group's change to remove the common trend
  
    Interactive · Difference-in-differences (illustrative numbers)
    Common trend (hits both groups)
      
    True effect of the policy
      
    Treated group was already drifting differently
      
    
    
    Treated group starts at 20, comparison at 24 (invented units). Dashed line: where the treated group would have gone without the policy, if it followed the comparison's trend.
  
  With a common trend of +1.5 and no true effect, the treated group still rises by 1.5, so a naive "before and after" would claim a policy success. The difference-in-differences correctly says zero. The design's weak point is the third slider: it assumes the two groups would have moved in parallel without the policy. If the treated group was already on a different path, the estimate is off by exactly that gap. Good papers show the two lines tracking each other for years before the policy.

  Instruments and cutoffs: two more ways to find a coin
  An instrumental variable is something that nudges the thing you care about but has no other route to the outcome. Angrist (1990) studied whether military service in the Vietnam era changed later earnings. Men who chose to serve differ from those who did not, so a plain comparison is confounded. But the draft lottery assigned numbers by birthday at random, and a low number made service much more likely. The lottery number is a coin that pushes people into uniform. Its effect on earnings, divided by its effect on service, isolates the effect of service for the people who served because of the lottery. His study found that service lowered earnings for the veterans concerned.
  effect of service = (earnings gap by lottery group) ÷ (service gap by lottery group)the "Wald" ratio: reduced form over first stage
  A worked number, invented to show the arithmetic: suppose 35% of low-number men served and 20% of high-number men, a first stage of 0.15; and low-number men earned $300 less a year on average. Then the estimated effect of serving is −300 ÷ 0.15 = −$2,000 a year. Dividing by a small first stage blows up noise, which is why weak instruments are dangerous.
  A regression discontinuity uses a cut-off. Israeli school classes are capped at 40 pupils under a rule attributed to the twelfth-century scholar Maimonides, so a school with 40 pupils has one class and a school with 41 splits into two small classes of about 20. Angrist and Lavy (1999) compared schools just either side of those thresholds, where enrolment is close to a coin flip, to estimate what class size does to test scores. Slide the window below: narrow windows are fairer but noisier.
  
    Interactive · Regression discontinuity (illustrative data)
    Window around the cut-off (± points)
      
    True jump at the cut-off
      
    
    
    400 invented students; those scoring 60 or more get a support programme. Outcome rises 0.3 per score point anyway, so just comparing everyone above with everyone below overstates the jump.
  

  Standard errors: how much would the answer wobble?
  Every estimate comes from one sample, and a different sample would have given a different number. The standard error is the typical size of that wobble. For a regression slope it shrinks with the square root of the sample size, grows with the noise around the line, and shrinks when x varies a lot:
  SE(b) ≈ (noise around the line) ÷ ( spread of x × √n )four times the data halves the wobble
  The rule of thumb: an estimate more than about two standard errors from zero is called statistically significant at the 5% level. Significant means "hard to produce by sampling wobble alone if the true effect were zero". It does not mean big or important, and it does not mean true. Run 300 studies of a real but modest effect below and see who gets stars.
  
    Interactive · Run 300 studies (illustrative simulation)
    Sample size per study
      
    True slope
      
    Run 300 new studies
    
    significant at 5%   not significant
    
  
  At n = 50 and a true slope of 0.3, a little over half of the studies reach significance, and those average about 0.40, a third above the truth, because only the lucky overestimates clear the bar. Drop to n = 10 and a true slope of 0.05 and the few stars (about 8%) average roughly 0.7, fourteen times too big. Published results are mostly the significant ones, so they exaggerate. That is one reason newer replications so often find smaller effects.

  P-hacking: test enough things and something will glow
  A test at the 5% level is wrong by chance 1 time in 20 when nothing is going on. Test 20 unrelated outcomes and the chance that at least one glows is 1 − 0.9520 = 64%. A researcher who tries several outcomes, subgroups or control sets and reports the one that "worked" is p-hacking, whether on purpose or just by wandering a garden of forking paths. Simmons, Nelson and Simonsohn (2011) showed how easy it is: using flexible, ordinary-looking choices they "found" that listening to a song about being older made students younger by about a year and a half. Economists are not exempt: Brodeur and co-authors (2016) found an unnatural pile-up of results just below the 0.05 line in top economics journals.
  
    Interactive · A study of nothing (illustrative simulation)
    Outcomes you test (all truly unaffected)
      
    Run one studyRun 500 studies
    
    
    Each outcome compares two groups of 50 drawn from the same distribution, so every "effect" is noise. A cell lights up when |t| passes the 5% cut-off (1.98).
  
  Defences: decide the analysis before seeing the data (pre-registration), report every test you ran, adjust the threshold when you run many, and above all prefer designs, like the ones above, where the answer does not depend on which controls you picked.

  What a graduate just sees
  
    A regression coefficient is a comparison between different people, not a lever you can pull.
    Before believing "X raises Y", ask what else moves with X and also moves Y. The bias is its effect times its link to X.
    Adding controls can only fix confounders you measured. A coin flip fixes the ones you did not.
    Every natural-experiment study is a hunt for a coin: a lottery, a border, a cut-off, a law that changed for some and not others.
    Difference-in-differences needs parallel trends; ask to see the years before the policy.
    "Significant" means hard to get from wobble alone. It says nothing about size, and a small-sample star probably overstates.
    If you can run twenty tests, one will light up. Ask how many were run.
  

  Where it shows up next
  
    Development Economics (34/48): randomised trials and natural experiments are how economists test what actually helps poor countries.
    Behavioral Economics (34/46): the replication record in that piece depends on this one, from effect sizes to p-hacking.
    Marketing Research & Analytics (34/59): "did the campaign cause the sales, or did summer?" is difference-in-differences.
    Machine Learning (34/32): prediction asks "what will happen"; econometrics asks "what would happen if I changed it". They need each other.
    In the world: central banks and tax agencies, tech companies' A/B test teams, public-health trials, and every think-tank report that says "causes".
  

  If you want to go deeper
  
    📘 Textbook. James Stock and Mark Watson, Introduction to Econometrics (Pearson). A standard undergraduate text, strong on causal thinking and applications.
    📘 Textbook. Jeffrey Wooldridge, Introductory Econometrics: A Modern Approach (Cengage). The other standard choice, more algebra and more worked data.
    📘 Gentler. Joshua Angrist and Jörn-Steffen Pischke, Mastering 'Metrics: The Path from Cause to Effect (Princeton, 2015). Written for people who are not specialists, built around exactly the designs above.
    🌐 Free course. MIT OpenCourseWare 14.32 Econometrics (Spring 2007, Joshua Angrist): undergraduate course with lecture notes and problem sets.
  

  
    The takeaway
    A regression gives you a comparison, and a comparison is only a cause if nothing else differs between the groups. Econometrics is a toolkit for making that true or nearly true (randomise, difference out a trend, use a lottery, zoom in on a cut-off), plus the humility to report how much your number would wobble and how many ways you tried.
    Try this: the next time a headline says "X linked to Y", name one confounder that could drive both, and ask what coin flip, border or cut-off could separate them.
  

  
    Sources & Further Reading
    Sources. Least squares, omitted-variable bias, standard errors and the Wald estimator are standard textbook material (Stock & Watson; Wooldridge; Angrist & Pischke). Card & Krueger, "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania," American Economic Review 84(4), 1994; the Nobel Prize in Economic Sciences 2021 went to David Card (half) and Joshua Angrist and Guido Imbens (half, "for their methodological contributions to the analysis of causal relationships"). Angrist, "Lifetime Earnings and the Vietnam Era Draft Lottery," 1990; Angrist & Lavy, "Using Maimonides' Rule to Estimate the Effect of Class Size on Scholastic Achievement," 1999; regression discontinuity was introduced by Thistlethwaite & Campbell in 1960. Angrist & Pischke, "The Credibility Revolution in Empirical Economics," Journal of Economic Perspectives 2010; Leamer, "Let's Take the Con out of Econometrics," 1983. Simmons, Nelson & Simonsohn, "False-Positive Psychology," 2011; Brodeur, Lé, Sangnier & Zylberberg, "Star Wars: The Empirics Strike Back," 2016. The 2019 economics Nobel went to Banerjee, Duflo and Kremer. Every number an interactive shows was computed in node from the code on this page: the best line is 47.0 + 3.53 × hours; omitted-variable bias at strength 1.2 is 8 × 1.2 ÷ (1.2² + 4) = 1.76 in theory (1.67 in this sample); 1 − 0.9520 = 64.2%. All datasets are invented.
    
      🌐 MIT OCW 14.32 Econometrics
      📘 Angrist & Pischke, Mastering 'Metrics
      📘 Stock & Watson, Introduction to Econometrics


Step 8. Game Theory: When Your Best Move Depends on Mine
--------------------------------------------------------
Why: When your best move depends on mine: prices, auctions, arms races.
Two shops pick prices, two countries pick tariffs, a hundred bidders pick bids. Game theory is the maths of choosing when everyone else is choosing too, and it explains why smart people can trap themselves.

Two petrol stations face each other across a road. If both charge a fair price, both make a decent profit. If one cuts its price, it steals the customers and makes more, for a while. So each is tempted to cut. Both cut. Both now earn less than before, and neither can raise prices alone without losing everyone. Nobody was stupid. Each did the sensible thing given what the other might do.
  That pattern, a trap built entirely out of sensible choices, is the first thing game theory teaches you to see. It then teaches how to get out of the trap: by repeating the game, by changing who moves first, by changing the rules. The same toolkit explains the design of radio-spectrum auctions and the matching of kidney donors.

  
    The course in one box
    
      Who takes itEconomics majors in year 2, straight after microeconomics. Business schools teach it inside strategy courses, and computer science and political science programmes teach it too, because algorithms and elections are games.
      What it is forPredicting what rational, self-interested players will do when each one's best move depends on the others, and designing rules so the result is a good one.
      Comes afterMicroeconomics (34/42).
      Leads toStrategy & Competition (34/54); it also underlies the auctions and pricing in Pricing (34/61) and the rules-of-the-game thinking in Public Economics (34/49).
      What this piece gives youYou will not be able to solve a multi-stage game with incomplete information. You will see why a dominant strategy can leave everyone worse off, why a threat only works if carrying it out is in your interest, and why an auction format changes how people bid. Easier first reads: the library's The Social Contract (18/14) asks why we accept rules and a state at all, and Anarchism (31/04) covers the commons problem in words. This piece puts numbers on that kind of trap.
    
  

  Players, strategies, payoffs, and the Nash equilibrium
  A game has three ingredients: the players (who decides), the strategies (what each can do) and the payoffs (what each gets for every combination of choices, in whatever they care about: money, time, votes, survival). Two players with two options each fit in a small grid, the payoff matrix. One player picks the row, the other the column, and each cell holds two numbers: the row player's payoff first, the column player's second.
  Two ideas solve most such grids. A dominant strategy is one that is best for you whatever the other does: if it exists, play it. When neither player has one, look for a Nash equilibrium: a pair of choices where neither player could do better by changing their own choice alone. John Nash proved in 1950 that every finite game has at least one, possibly in mixed strategies (rolling dice between options). Nash shared the 1994 economics Nobel prize with John Harsanyi and Reinhard Selten for it.
  Nash equilibrium: each player's choice is a best response to the other'snobody gains by switching alone; it says nothing about whether the outcome is good
  
    Interactive · The payoff-matrix workbench
    
      Prisoner's dilemma
      Stag hunt
      Chicken
      Matching pennies
      Which side to drive
    
    
    
    
    Each cell: your payoff (row player), then the column player's. Edit the numbers; the analysis updates. Shaded = a best response for that player; outlined cell = Nash equilibrium.
  
  Without JavaScript: in the prisoner's dilemma (5, 3, 1, 0), defecting is dominant for both players, the only Nash equilibrium is both defecting for 1 each, and both cooperating would give 3 each.
  Try the presets and read the verdicts. Chicken and the driving game have two pure equilibria and a mixed one at 50:50; matching pennies has no pure equilibrium at all, only the mixed one, which is why a goalkeeper and a penalty-taker must be unpredictable. In matching pennies, if the row player leans even slightly toward heads, the column player exploits it.

  The prisoner's dilemma: why smart players choose badly
  Merrill Flood and Melvin Dresher devised it at the RAND Corporation in 1950, and Albert Tucker gave it its story: two suspects, questioned separately. Stay silent (cooperate) and both get a light sentence. Betray the other (defect) and you walk free while he gets the worst. Both betray and both get a heavy sentence. With points instead of years, the standard payoffs are: temptation 5, mutual cooperation 3, mutual defection 1, being betrayed 0.
  Whatever the other does, defecting pays you more: 5 beats 3 if she cooperates, 1 beats 0 if she defects. So both defect and both get 1, when both cooperating would have given 3 each. The equilibrium is stable and bad. That is the logic of arms races, overfishing, price wars and the two petrol stations. Now play it. Ten rounds against a computer opponent whose rule you can pick; you know the game is exactly ten rounds long.
  
    Interactive · Play ten rounds
    
      Tit for tat
      Always defect
      Always cooperate
      Grudger
      Coin flip
    
    CooperateDefectRestart
    
    Payoffs per round: both cooperate 3 each; you defect on a cooperator 5 to 0; both defect 1 each. Tit for tat copies your last move; the grudger cooperates until you defect once, then never again.
  
  Against tit for tat, the best ten-round total is 32: cooperate nine times, defect on the last round when it can no longer be punished. Against always-defect, nothing beats 10. Against always-cooperate you can take 50. The right play depends on who is on the other side, which is the whole subject in one sentence.

  Repeated games: when tomorrow changes today
  Meet the same person again and again and the logic changes: defecting today invites revenge tomorrow. In 1980 the political scientist Robert Axelrod invited game theorists to send in programs for a repeated prisoner's dilemma, each program playing every other for 200 moves. The winner, in this tournament and in a second, larger one, was the simplest entry: tit for tat, sent by the psychologist and game theorist Anatol Rapoport. It cooperates first, then copies the opponent's last move. Axelrod's 1984 book The Evolution of Cooperation drew the lesson: successful strategies were nice (never defect first), provocable (punish defection at once), forgiving (return to cooperation) and clear (easy to read).
  Tit for tat never beats the partner it faces; the best it can do in a single match is tie. It won by getting good scores out of nearly everyone. Run your own small tournament. Six kinds of players meet in a 200-round round robin; add copies of Always-defect and see whether the winner changes.
  
    Interactive · A round-robin tournament (a toy field)
    Always-defect entrants
      
    
      No mistakes
      5% of moves slip
    
    
    
    Entrants: Always cooperate, Tit for tat, Grudger, Pavlov (cooperate only if both did the same last round), Coin flip, plus the Always-defect copies. Everyone meets everyone else for 200 rounds; bars show the average points per round. Payoffs 5, 3, 1, 0. A toy field, far smaller than Axelrod's.
  
  With no Always-defect, the nice, retaliating strategies fill the top. Add more defectors and the field gets harsher: the defectors climb to the top once enough naive cooperators and coin-flippers are there to be exploited, while their scores against each other stay at 1. Success in a repeated game depends on who else is playing. Turn on the 5% slips and tit for tat loses ground too: after one accidental defection two tit-for-tat players can lock into a spiral of mutual retaliation. Later work found more forgiving variants, such as generous tit for tat, and the Pavlov rule do better in noisy worlds. Cooperation without a referee needs a future worth protecting, which is why long-term relationships, reputations and repeated dealings support trust that one-off meetings cannot.

  Coordination games: when you just need to match
  Not every game is a conflict. In the driving game, both of you want to pick the same side; which side hardly matters. Load "Which side to drive" above: two equilibria, both good, and nothing in the payoffs says which. Conventions (driving on the left in the UK, Japan and India, on the right in most of the rest of the world) are equilibria picked by history. The stag hunt adds risk: both hunting stag gives 4 each, but hunting hare alone is safe at 3. Two equilibria again, one better but riskier. Thomas Schelling, in The Strategy of Conflict (1960), noticed that when payoffs cannot choose, players use focal points: the obvious, salient option, like meeting "at the clock" in a station. Many real-world failures, from bank runs to standards wars, are coordination games stuck on the bad equilibrium.

  Sequential games: the credible threat
  When players move in turn, you solve the game backwards: start at the end, see what the last mover would do, and work back. Take a company deciding whether to enter a market held by an incumbent. The incumbent says: "If you enter, I'll start a price war." Should the entrant believe it? Only if fighting is actually best for the incumbent after entry happens. If, once the entrant is in, the war would cost the incumbent more than sharing the market, the threat is empty. A rational entrant sees that and enters. Reinhard Selten's refinement of Nash equilibrium, subgame-perfect equilibrium, rules such empty threats out.
  
    Interactive · Entry, threats and commitment
    Incumbent's profit if it fights after entry
      
    
      No commitment
      Commit to fight (contract penalty)
    
    
    
    Payoffs (entrant, incumbent): stay out (0, 10); enter and incumbent accommodates (4, 5); enter and incumbent fights (−2, your slider). The commitment option makes accommodating cost the incumbent 3 (so it pays only 2).
  
  The slider has a threshold. With a war worth less than 5 to the incumbent, the threat is empty and the entrant walks in. Raise the war's payoff above 5, or use the commitment button, and the threat becomes credible, the entrant stays out, and the incumbent never has to fight. Commitment devices (spare factory capacity already built, a contract that punishes backing down, a reputation) have this strange property: by taking away your own options they can make you better off.

  Auctions: bid your value, or shave it
  An auction is a game where the rules decide the payoffs. In a sealed-bid second-price auction, the highest bidder wins but pays only the second-highest bid. William Vickrey analysed it in 1961 (it had been used by stamp collectors since 1893). Here, bidding exactly your true value is a dominant strategy: bidding more risks paying above your value, bidding less risks losing an item you would have profited from. In a first-price auction you pay what you bid, so you shave your bid below your value, trading a bigger profit if you win against a smaller chance of winning. With four bidders whose values are equally likely to be anywhere in a range, the equilibrium bid is three-quarters of your value (in general, (n−1)/n of it).
  
    Interactive · Four bidders, two formats (3,000 simulated auctions)
    Your bid as a share of your value
      
    
    second-price   first-price
    
    Values for all four bidders are drawn uniformly between 0 and 100. The three rivals bid their full value in the second-price format and three-quarters of it in the first-price format. Your average profit is per auction, including those you lose.
  
  Slide the share. In the second-price format your profit peaks at 100% of your value; in the first-price format it peaks near 74%. Also watch the seller's average revenue: it is about 60 in both formats, a first glimpse of the revenue equivalence theorem. Real auctions differ from this textbook case. When bidders are unsure of the item's true value (oil leases, spectrum) the winner tends to be the one who overestimated most, the winner's curse.

  Mechanism design: design the game, then let people play it
  Game theory asks: given the rules, what will players do? Mechanism design turns it around: what rules would make self-interested players do what we want? Two famous cases.
  Spectrum auctions. Governments own radio frequencies and need to hand them to phone companies. For decades they used beauty contests and lotteries. From July 1994 the US Federal Communications Commission instead ran a "simultaneous multiple round" auction, where many related licences are sold together over many rounds and bidders see each other's bids. Theorists including Paul Milgrom and Robert Wilson helped shape the design, and they shared the 2020 economics Nobel prize for improvements to auction theory and new auction formats. In 2000 the UK's 3G auction raised £22.5 billion from five licences.
  Kidney exchange. Many people who need a kidney have a willing donor who is the wrong blood type. Selling kidneys is illegal almost everywhere, so prices cannot match people. Alvin Roth, Tayfun Sönmez and Utku Ünver designed algorithms (their 2004 paper in the Quarterly Journal of Economics) in which pair A's donor gives to pair B's patient while B's donor gives to A's: a swap, extended to longer cycles and to chains started by an altruistic donor. The first paired exchange recorded was in South Korea in 1991 and the first in the US in 2000; by 2012 the US National Kidney Registry had organised a 60-surgery chain, and by 2014 one of 70. Roth shared the 2012 economics Nobel prize with Lloyd Shapley for the theory of stable allocations and the practice of market design.
  Game theory predicts what people do under the rules. Mechanism design chooses the rules.

  What a graduate just sees
  
    Before asking what someone will do, ask what they would do if they expected you to do each of your options.
    A dominant strategy needs no guess about the other player. If it exists, it settles the game.
    A Nash equilibrium is stable, not good. The prisoner's dilemma's equilibrium leaves both worse off than cooperating.
    Repetition changes the game: cooperation can be rational when the future is long enough and defection is punishable.
    A threat counts only if carrying it out is in your interest when the time comes. Otherwise build a commitment or expect to be called.
    When payoffs do not choose between equilibria, the focal point does: salience, convention, a shared story.
    In a second-price auction, bid your value. In a first-price auction, shave your bid. The rules decide how people behave.
    When a market cannot use prices, you can still design the rules.
  

  Where it shows up next
  
    Strategy & Competition (34/54): firms setting prices, building capacity and deciding whether to enter are playing the games above.
    Pricing (34/61): discounts, price matching and surge pricing are games with customers and competitors.
    Behavioral Economics (34/46): the next piece asks when real people do not play the equilibrium, as in the ultimatum game.
    Machine Learning (34/32) and Computer Networks (34/27): game-playing AIs and congestion control both apply these ideas.
    In the world: online-advertising auctions, spectrum sales, school and hospital matching, climate treaties, arms control.
  

  If you want to go deeper
  
    📘 Textbook. Robert Gibbons, Game Theory for Applied Economists (Princeton, 1992). Short and widely used in economics programmes.
    📘 Textbook. Martin Osborne, An Introduction to Game Theory (Oxford). A fuller undergraduate treatment with many worked games.
    📘 Gentler. Avinash Dixit and Barry Nalebuff, The Art of Strategy (Norton, 2008), and Robert Axelrod, The Evolution of Cooperation (1984). Real cases, almost no algebra.
    🌐 Free course. Open Yale Courses ECON 159: Game Theory (Ben Polak, 24 lectures, recorded Fall 2007), covering Nash equilibrium, backward induction and signalling.
  

  
    The takeaway
    When your best move depends on others' moves, look for dominant strategies, then for equilibria, and remember that a stable outcome can be a bad one. Repetition, commitment and well-designed rules are the three ways out, and each is a way of changing the game rather than the players.
    Try this: this week, spot a situation where you and someone else are each waiting for the other to move first or to give in, and ask what the payoff grid looks like and whether it is a game you can change.
  

  
    Sources & Further Reading
    Sources. Dominance, Nash equilibrium, mixed strategies, backward induction and the second-price result are standard textbook game theory (Gibbons; Osborne; Polak's Yale lectures). The prisoner's dilemma was devised by Flood and Dresher at RAND in 1950 and named by Tucker (Wikipedia, "Prisoner's dilemma"). Nash, "Equilibrium points in n-person games," PNAS 36, 1950; the 1994 economics Nobel prize went to Nash, Harsanyi and Selten. Axelrod's tournaments (c. 1980) and 1984 book: tit for tat, by Anatol Rapoport, won both; matches were 200 moves; the death-spiral under noise and generous tit for tat are as described in Wikipedia's "Tit for tat". Schelling, The Strategy of Conflict (1960). The second-price auction was described by Vickrey in 1961 and used by stamp collectors from 1893. The FCC has held spectrum auctions since July 1994 using a simultaneous multiple round design; the UK's 2000 auction raised £22.5 billion from five 3G licences (Wikipedia, "Spectrum auction"). The 2020 prize went to Milgrom and Wilson, the 2012 prize to Roth and Shapley. Kidney exchange history from Wikipedia's "Kidney paired donation"; Roth, Sönmez & Ünver, "Kidney Exchange," QJE 2004. Every number an interactive shows was computed in node from the code on this page: ten-round best score against tit for tat 32, against always-defect 10, against always-cooperate 50; the four-bidder first-price optimum is near 0.74 (theory 0.75); the average winning price is near 60 in both formats; the toy tournament's ranking depends on the chosen field and is not Axelrod's result.
    
      🌐 Open Yale Courses ECON 159 Game Theory
      📘 Dixit & Nalebuff, The Art of Strategy
      📘 Axelrod, The Evolution of Cooperation


Step 9. Public Economics: What Governments Can and Can't Fix
------------------------------------------------------------
Why: Taxes, public goods and inequality: what governments can and can't fix.
Why tax at all, who really pays it, what redistribution costs, how inequality is measured, and why the state fails in its own ways.

Look at your last pay slip, or at a shop receipt. Part of every price you pay and every wage you earn goes to a government, and the road you walked on, the water you drank and the vaccine your grandparents took were paid for that way. In Denmark that share is about 45% of what the whole economy produces; in Mexico, 18%. In Brazil the richest tenth gets close to 60% of income; in Sweden, under 30%. Nobody voted to make these numbers exactly what they are, yet they shape everything from how long you wait for a doctor to whether your child can climb out of poverty.
  A public economics graduate learns to ask two questions of any policy. Does it fix a real failure of markets, and who ends up paying and gaining once everybody has reacted? The first is a question about efficiency, the second about fairness, and the discipline's honesty comes from keeping them separate and keeping a third, awkward one in view: governments fail too.

  
    The course in one box
    
      Who takes itEconomics students as a field course in year 2 or 3 (MIT lists it as 14.41, Public Finance and Public Policy), and public-policy schools as part of the core. Business students meet its tax half in their finance courses.
      What it is forJudging what governments should do in an economy, what taxes and benefits really do to people, and what they cost in lost output.
      Comes afterMicroeconomics (34/42): surplus, elasticity and externalities do all the work.
      Leads toNo subject in the library requires it yet. It is the policy end of the Economics major, next to Macroeconomics (34/43) and Development Economics (34/48).
      What this piece gives youYou won't be able to design a tax code. You will see why who writes the cheque is not who pays, why cutting inequality and growing the pie can pull against each other, and how much richer countries actually redistribute. Easier first reads: How Taxes Actually Work (03/07), Where Public Money Goes (18/08), The Mixed Economy (31/05) and Climbing the Ladder: What the Data Says About Social Mobility (30/04).
    
  

  Why governments are in the economy at all: public goods
  Markets are good at things that can be bought, one buyer at a time. They are bad at things whose benefits spill over to everyone. A public good is non-excludable (you can't keep non-payers out) and non-rival (my use doesn't reduce yours): clean air, a lighthouse, defence, a vaccination campaign. Each person would prefer that someone else pays, and the result is that a private market supplies too little. This is the free-rider problem, and it is the oldest justification for taxation: a way to make everyone pay for something everyone wants.
  The game below is the classic experiment. Four players each hold 20 tokens. Whatever goes into the shared pot is multiplied by 1.6 and split equally, so each token you give returns 0.4 to you and 0.4 to each of the others.
  
    Interactive · The shared pot
    You give
      
    The others give, on average
      
    
    
    Tokens = money. Payoff = 20 − your gift + 0.4 × (all gifts). Invented game parameters.
  
  Whatever the others do, you always do better by giving nothing: every token you give costs you 1 and returns 0.4. But if all four give nothing, each ends with 20, while if all four give everything, each ends with 32. The best private choice and the best group result point in opposite directions. In real experiments people usually start by giving around half, and contributions drift down over repeated rounds, though a core of givers remains; allowing players to punish free riders keeps contributions higher. Governments do the "punishment": they make paying compulsory. Elinor Ostrom (Nobel 2009) added that communities can often solve the problem themselves, so tax is one solution, not the only one.

  Insurance and information: why some markets need a mandate
  Insurance is a market for sharing risk. It can fall apart when buyers know more about their own risk than the seller does: adverse selection, the problem George Akerlof described for used cars in 1970 and which won him a share of the 2001 Nobel Prize. If the insurer charges the average price, the low-risk customers find it too dear and leave; the pool gets sicker; the price rises; more leave. Governments respond with mandates (everyone must be in), public insurance, or subsidies.
  
    Interactive · The insurance spiral
    
      Voluntary
      Everyone must join
    
    How much people value cover relative to their expected cost
      
    
    
    Illustrative: 1,000 people with expected yearly costs from 0 to 999, evenly spread. Each will pay up to the multiple times their own expected cost. The premium is the average cost of those who buy.
  
  If people value cover at exactly their expected cost (multiple 1), a voluntary market collapses to almost nobody. At 1.3 the premium settles at 812 and only 37.5% buy. At 2 it settles at about 666 and two thirds buy. Even a keen market never covers everyone. A mandate fixes the premium at the pool average, 500, and covers everyone, but it forces in those who value cover at less than that price: 38% at multiple 1.3. Mandates trade a market failure for a compulsion, which is why the argument over them (Switzerland and the Netherlands require everyone to buy private health cover; the UK and Canada fund care from taxes instead) is fierce.

  Who really pays a tax
  The law says who must send the money. Tax incidence asks who ends up poorer. Put a tax on sellers and they raise the price; put it on buyers and they bid less. In both cases the price buyers pay and the price sellers keep move apart by the tax, and the split depends only on who can walk away more easily. The side that is less responsive to price carries more of the burden.
  share paid by buyers = supply elasticity ÷ (supply elasticity + demand elasticity)elasticities in size. Same answer whoever the law names. Short: the stickier side pays
  
    Interactive · A 10-unit tax on a 100-unit good
    
      Law names the sellers
      Law names the buyers
    
    Buyers' price sensitivity (demand elasticity)
      
    Sellers' price sensitivity (supply elasticity)
      
    
    
    Small-tax approximation in a linear market. The 10 and 100 are invented.
  
  With demand elasticity 0.5 and supply 1.5 buyers pay 75% of the tax, whichever side the law names. That is why economists are suspicious of a payroll tax "paid half by the employer": the split on the form is a legal fiction, and the usual finding is that workers bear most of it through lower wages. And it is why a tax on something people cannot easily give up, like fuel in the short run, raises revenue and mostly falls on the buyer.

  Efficiency and equity: the leaky bucket
  Taxes that move money from richer to poorer lose some on the way, through administration and, more importantly, because people change what they do. The economist Arthur Okun called this the leaky bucket (1975): carry a bucket of money from rich to poor and some leaks out. How big is the leak? It depends on how much people cut work, move income or avoid tax when the rate rises, which economists summarise as the elasticity of taxable income, e. In a simple model where the tax base shrinks by (1 − t)e, revenue t(1 − t)e peaks at a rate of 1 ÷ (1 + e): beyond that, a lower rate would collect more. This is the Laffer curve.
  
    Interactive · How high can a rate go?
    Tax rate on extra income
      
    How strongly people respond (e)
      
    
    
    Textbook toy: taxable income = (1 − t)e. Real estimates of e vary a lot between people, countries and time horizons, and are debated.
  
  With e = 0.5 revenue peaks at a rate of 67%; with e = 1, at 50%. Most of the argument is not about the curve, which almost nobody denies, but where real countries sit on it and how large e is for high earners. Equity-minded economists answer that a leak does not mean the bucket should stay empty; efficiency-minded ones that growth is the best antipoverty tool. Both statements are true. The disagreement is over weights and measured responses, so ask for both.

  Inequality, and how it is measured
  Inequality is easy to feel and hard to summarise. The standard picture is the Lorenz curve: line everyone up from poorest to richest and plot the share of income held by the bottom 10%, 20%, and so on. Perfect equality is a straight diagonal; the more it sags, the less equal. The Gini coefficient (Corrado Gini, 1912) is the area between the diagonal and the curve, doubled, so that 0 means everyone has the same and 1 means one person has everything. It is one number, and it hides shapes: two very different distributions can have the same Gini. For example, a society where half the people have nothing and the other half share equally has a Gini of 0.5, exactly like one where the bottom three quarters hold a quarter of the income and the top quarter three quarters. Top shares tell the other half of the story.
  
    Interactive · Build a Lorenz curve
    
      Equal
      Sweden / Norway-like
      India
      South Africa
    
    Share of income held by the bottom 50%
      
    Share of income held by the top 10%
      
    
    
    Three-slice approximation (bottom 50%, middle 40%, top 10%), so it understates the true Gini. The default is the global distribution in the World Inequality Report 2026 (top 10% about 53%, bottom half about 8%, pre-tax income); presets are shares reported for those countries in the same report, rounded.
  
  The World Inequality Report 2026 (Dec 2025) puts the global top 10% at about 53% of income and the bottom half at about 8%. By country, it reports South Africa's top 10% near 66% and Brazil, Mexico, Chile and Colombia near 60%, India, Thailand and Türkiye above 50%, and Sweden and Norway under 30%. A caution on the other measure: the World Bank's Gini for South Africa is 54.1 (2022), Brazil 50.3 (2024), Mexico 42.6 (2024), the United States 41.8 (2024), Germany 33.7 (2022), Sweden 29.3 (2023), Slovenia 24.7 (2023). These measure income or consumption per person from household surveys, which tend to miss the very top; the report's figures use tax and national accounts to see more of it. The two kinds of number will not agree and neither is wrong.

  The welfare state across countries
  The modern welfare state grew from Bismarck's social insurance laws in 1880s Germany and from Britain's 1942 Beveridge Report, which named five "giants" to slay: want, disease, ignorance, squalor and idleness. Today governments differ in how large they are and in how they spend: through cash for pensions and the unemployed, services like health and school, or tax breaks. The OECD (rich-country club) measures the size of the state's take as tax revenue as a share of GDP: 34.1% on average in 2024, from 18.3% in Mexico to 45.2% in Denmark.
  The same OECD database measures inequality before taxes and transfers ("market income") and after ("disposable income"). The gap is what the tax-and-benefit system does. Pick a country.
  
    Interactive · What does the state change?
    
    
    
    Tax revenue incl. social contributions, % of GDP, 2024 (OECD Revenue Statistics 2025). Gini from the OECD Income Distribution Database, latest year available (2022 to 2024), whole population. The OECD treats public pensions as transfers, not market income, so ageing countries show larger cuts.
  
  Nordic and continental European countries start with market-income Ginis around 0.43 to 0.52 and end at 0.28 to 0.33, a cut of a third or more. The United States starts about as unequal as the UK and France (0.51) and ends at 0.39. Mexico barely moves: 0.416 to 0.400, a cut of 4%, consistent with its small tax take. Notice also that the rich-country spread is mostly in how much they redistribute, not in how unequal the market leaves them. The sociologist Gøsta Esping-Andersen's Three Worlds of Welfare Capitalism (1990) sorted rich states into liberal, conservative and social-democratic types; most scholars now use more families, and regard the groups as tendencies.

  Government failure: the same logic, pointed the other way
  If markets can fail, so can the people who fix them. Public choice economics (James Buchanan, Nobel 1986, and Gordon Tullock) applies the same assumptions to voters, politicians and civil servants as to shoppers: they respond to incentives. Mancur Olson's The Logic of Collective Action (1965) explained why small groups with a lot at stake organise better than large groups with a little each. A subsidy that gives each of 50 firms 2 million and costs each of 100 million taxpayers 1 will be lobbied for by the firms and ignored by the taxpayers (illustrative numbers). Anne Krueger called the effort spent winning such favours rent-seeking (1974).
  Other failures are about knowledge: planners seldom know what local people know; programmes outlast their purpose; and well-meant rules cause side effects. The sound reply is not "government never works", which smallpox (declared eradicated by WHO in 1980) refutes, but Harold Demsetz's warning against the "nirvana fallacy" (1969): comparing a real market with an ideal government, or the reverse, instead of one real option with another. A graduate asks, for each policy, who gains, who pays, who is watching, and what happens after the first round of reactions.

  What a graduate just sees
  
    Public goods go unsupplied because everyone prefers to free-ride; compulsion or good rules fix that.
    When buyers know less than sellers, or the reverse, markets can unravel. Insurance is the case study.
    The law says who sends the cheque; elasticities say who pays. The less flexible side bears more.
    Redistribution has a leak, and the size of the leak is an empirical question, not an ideology.
    A Gini is one number. Ask for top shares, the bottom half and the time frame.
    Rich countries differ less in market inequality than in how much their taxes and benefits change it.
    Governments fail from incentives, not just stupidity: concentrated gains beat diffuse costs.
    Compare real policies, not a flawed market with a perfect state or the reverse.
  

  Where it shows up next
  
    Macroeconomics (34/43): the budget deficit, the debt ratio and the multiplier are this subject at the scale of the whole economy.
    Development Economics (34/48): why many poorer countries collect far less tax (Nigeria 8% of GDP, Kenya 16%, Mexico 18%, in the OECD's global series) and what that does to schools and roads.
    Behavioral Economics (34/46): defaults and nudges in pension and tax design.
    In the world: a sugar tax, a carbon price, a minimum pension, a budget speech, the debate over a wealth tax or a universal income.
  

  If you want to go deeper
  
    📘 Textbook. Jonathan Gruber, Public Finance and Public Policy (Worth/Macmillan). The usual undergraduate text, strong on evidence from natural experiments.
    📘 Textbook. Harvey Rosen and Ted Gayer, Public Finance (McGraw-Hill). Another standard choice, a bit more theory.
    📘 Gentler. Anthony Atkinson, Inequality: What Can Be Done? (Harvard University Press, 2015), or Branko Milanovic, Global Inequality (Harvard University Press, 2016): clear on the data, with proposals set out for and against.
    🌐 Free course. MIT OpenCourseWare 14.41 Public Finance and Public Policy (Fall 2024). Also World Inequality Report 2026, free online.
  

  
    The takeaway
    Governments belong in an economy where prices miss something: public goods nobody will buy alone, insurance that unravels, spillovers that land on others. Who writes the cheque for a tax is not who pays it. Redistribution has a leak that is real and measurable, inequality has to be measured with more than one number, and the state is subject to the same incentives it tries to correct.
    Try this: this week, take one tax or subsidy you have heard of and write down who legally pays, who you think really pays, and which side of the market could walk away more easily.
  

  
    Sources & Further Reading
    Sources. Concepts: standard textbook public finance (Gruber; Rosen and Gayer). OECD, Revenue Statistics 2025 (December 2025): OECD average 34.1% of GDP in 2024 (an estimate: unweighted average change for 36 countries applied to 2023), range 18.3% (Mexico) to 45.2% (Denmark); countrywide figures here are 2024 values for France 43.5, Italy 42.8, Sweden 41.4, Germany 38.0, UK 34.4, Brazil 33.7, United States 25.6, Korea 25.3, Chile 20.5; Kenya 15.6 (2024) and Nigeria 8.2 (2023) from the OECD's global revenue series. OECD Income Distribution Database, Gini before and after taxes and transfers: Denmark 2022 0.442 / 0.276; France 2023 0.517 / 0.299; Italy 2023 0.509 / 0.325; Sweden 2024 0.431 / 0.289; Germany 2023 0.497 / 0.307; UK 2023 0.522 / 0.367; Brazil 2022 0.523 / 0.451; Korea 2023 0.392 / 0.323; US 2023 0.506 / 0.394; Chile 2022 0.491 / 0.448; Mexico 2022 0.416 / 0.400. World Bank Poverty and Inequality Platform Gini (SI.POV.GINI), latest, accessed November 2026. World Inequality Report 2026 (World Inequality Lab, December 2025), headline shares as reported by Al Jazeera, 10 December 2025 (global top 10% 53%, bottom half 8%; South Africa 66% and 6%; Sweden and Norway top 10% under 30%, bottom half about 25%) and by Indian press for India (top 10% about 58%, bottom half about 15%); check the report itself for exact figures. Gini definition and the same-Gini example: Corrado Gini 1912, as described on Wikipedia. Mexico's tax, public goods experiments, Fehr and Gächter 2000: summary in Wikipedia articles. Akerlof, "The Market for 'Lemons,'" QJE 1970 (Nobel 2001); Ostrom, Governing the Commons (1990; Nobel 2009); Buchanan Nobel 1986; Olson 1965; Okun, Equality and Efficiency (1975); Krueger 1974; Demsetz 1969; Beveridge Report, published 1 December 1942 (some sources date it November 1942); Bismarck's social insurance in the 1880s; Esping-Andersen 1990; WHO declared smallpox eradicated in 1980. Computed in node: public goods game payoffs 20 (nobody gives) and 32 (all give); insurance equilibrium 812 / 37.5% at 1.3, 666 / 66.7% at 2, forced-in share 50 ÷ multiple; incidence shares es ÷ (es + ed); Laffer peak 1 ÷ (1 + e) = 67% at 0.5, 50% at 1; Ginis from the three-slice curve: global 0.59, South Africa 0.68, India 0.56, Sweden/Norway 0.32.
    
      🌐 MIT OCW 14.41 Public Finance and Public Policy
      🌐 World Inequality Report 2026
      📘 Gruber, Public Finance and Public Policy
      📘 Atkinson, Inequality: What Can Be Done?


Step 10. Behavioral Economics: How People Actually Choose
---------------------------------------------------------
Why: Real people aren't perfectly rational, and the ways they deviate are predictable.
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 11. International Trade & Finance: Why Countries Swap, and Why Currencies Swing
------------------------------------------------------------------------------------
Why: Why countries trade, why currencies move, and who wins and loses from both.
Trade makes countries richer on average and leaves particular people worse off. Currencies let money cross borders, and fixing one carries a price you cannot dodge. Two ideas, one subject.

Look at the label on whatever you are wearing. The cotton may have been grown in one country, spun in a second, sewn in a third and sold in a fourth, and the price you paid moved a little this morning because a currency did. A Big Mac costs about 6.22 US dollars in the United States and about 9.04 dollars in Switzerland, in the Economist's July 2026 edition of its index, though it is the same sandwich made by the same company.
  An economist looks at all of this and asks two questions. Why do countries trade at all, and who gains and who loses when they do? And why do prices in different currencies drift apart, and what happens when a government tries to hold its currency still? The first is trade theory. The second is international finance. Together they are the course.

  
    The course in one box
    
      Who takes itEconomics majors, usually as a year-3 field course; international business and finance programmes; and, in Business Administration, as a year-4 subject. Programmes usually split it in two: a trade half (comparative advantage, tariffs, gains and losses) and a finance half (exchange rates, balance of payments, crises). Doctoral courses such as MIT's 14.581 go much further into the evidence.
      What it is forTelling you who gains and who loses when borders open or close, and why a currency can hold steady for years and then break in a week.
      Comes afterMicroeconomics (34/42) and Macroeconomics (34/43).
      Leads toDevelopment Economics (34/48), where trade policy is one of the oldest arguments about how poor countries get rich, and Strategy & Competition (34/54), where firms decide where to make things and sell them.
      What this piece gives youYou will not be able to forecast an exchange rate (nobody can, reliably). You will see why trade lifts a country's total income, why the winners could pay the losers but often do not, why a tariff is a tax on your own consumers, and why you cannot have a fixed currency, free money flows and your own interest rate all at once. Easier first reads: Capitalism (31/01) for markets in general and Social Mobility (30/04) for what job loss does to a region. This one puts numbers on the trade-offs. This is how things work, not investment, legal or tax advice.
    
  

  Comparative advantage: trade pays even when you are worse at everything
  The surprising result in the subject is also the oldest. In 1817 David Ricardo argued, in his Principles of Political Economy and Taxation, that two countries both gain from trade even if one is better than the other at making everything. His example was English cloth and Portuguese wine. By his figures, Portugal could make a unit of wine with 80 labour-years and a unit of cloth with 90. England needed 120 for the wine and 100 for the cloth. Portugal is better at both.
  The trick is to count in opportunity cost, not hours. Every wine Portugal makes means giving up 80 ÷ 90 = 0.89 of a cloth. Every wine England makes means giving up 120 ÷ 100 = 1.2 cloths. Wine is cheaper, in what you give up, in Portugal. So Portugal should lean to wine and England to cloth, even though England is worse at both. Your lawyer may type faster than her assistant and still should not be typing. Time spent typing is time not spent on something worth more.
  gain if the price of a wine lies between the two opportunity costs0.89 cloth < price of one wine < 1.2 cloth: inside that band both countries are better off
  
    Interactive · Ricardo's calculator
    Each country has 1,000 labour-years and wants equal amounts of wine and cloth (a deliberately simple taste). Without trade each makes both. With trade each specialises where its opportunity cost is lower and swaps at the price you set.
    World price: cloth per 1 wine
      
    
    
    Ricardo's 1817 labour figures. Portugal with trade: 12.5 wine, sells it at your price. England: 10 cloth. Each ends up consuming equal amounts of both. Try 0.70 and 1.40: outside the band one country loses and would refuse to trade.
  
  The graduate's habit is to ask, before any claim about trade, "cheaper compared with what?" Comparative advantage is not about who is best. It is about what each side gives up.

  Gains from trade, and who loses
  Ricardo's model has one factor and no people. Real countries have farmers and teachers, owners and workers. When a country opens to imports of a good, the price falls to the world price. Buyers win, because the price is lower. Domestic makers lose, because they now sell at that lower price. In the standard supply-and-demand picture the gain to buyers is larger than the loss to makers, so the country as a whole is ahead. If the world price is above the home price the same logic runs the other way: makers win, buyers lose, and the total is again positive.
  
    Interactive · Open the border
    An illustrative market: demand is Q = 100 − P, supply is Q = P − 20, so with no trade the price is 60 and the quantity is 40. Move the world price and watch buyers' surplus (blue), makers' surplus (gold) and the net.
    World price
      
    
    
  
  The line to remember is "the winners could compensate the losers and still be ahead." Could. Whether they do is politics, not the model. Trade theory is a statement about the total. The diagram is silent about whose street the factory was on, which is where the evidence in the next section starts.

  The China shock: a gain that landed unevenly
  China joined the World Trade Organization in December 2001, and its exports of manufactured goods surged. Economists David Autor, David Dorn and Gordon Hanson compared US local labour markets by how much of their industry competed with Chinese imports. In a 2013 paper, "The China Syndrome", they found that places more exposed had more unemployment, lower labour-force participation and lower wages, with a rise in welfare transfers. They attributed, conservatively, about one quarter of the fall in US manufacturing employment between 1990 and 2007 to Chinese import competition.
  A 2016 follow-up, "The China Shock", found that adjustment was far slower than the textbook expects: wages and participation stayed depressed, and exposed workers faced more job churning and lower lifetime income. Their headline estimate is on the order of 2 million US jobs lost to Chinese imports in the years after 2000; the exact figure and its interpretation are disputed, and critics argue that the jobs lost in manufacturing alone are much smaller and that cheaper goods and new export jobs offset part of it.
  
    QuestionWhat the evidence says
    Did trade raise total income?Most economists say yes: cheaper inputs and goods for everyone, and more exports for some.
    Did everyone gain?No. Exposed US regions lost jobs and income, and recovery was slow.
    Is it the same everywhere?No. A study of German labour markets (Dauth, Findeisen and Suedekum, 2014) reports net employment gains there, because German firms sold into the same new markets.
    Where do economists disagree?On how large the US job losses were, how much was trade versus automation, and what policy should have done.
  
  Free trade is not "everyone wins". It is "enough is won that nobody had to lose, if someone pays".

  Tariffs: who actually pays
  A tariff is a tax on an imported good. The politician's picture is that the foreign seller pays. The economist's first question is "who ends up with the higher price, and who is paid less?" That is incidence, and it does not depend on who writes the cheque. If the country is a small buyer in a big world market, the foreign seller will not cut its price, so the domestic price rises by the whole tariff. If the buyer is large, the seller may cut its pre-tax price to keep the sale, and some of the tax lands abroad.
  Recent history put this to a test. In a 2019 paper on the 2018 US tariffs, Mary Amiti, Stephen Redding and David Weinstein found "complete passthrough" into US import prices: foreign sellers did not cut their prices, and US buyers bore the cost, about $1.4 billion a month of lost real income by the end of 2018. The tariffs from 2025 onward are a much larger test, and the research on them was still arriving when this was written.
  
    Interactive · Put a tariff on it
    Same market, world price fixed at 40. A tariff of t per unit; the pass-through is the share of it that shows up in the domestic price (100% = the foreign seller does not budge).
    Tariff per unit
      
    Pass-through to domestic price
      
    
    
    Net welfare = buyers' loss + makers' gain + tariff revenue. At full pass-through it is always negative (the "deadweight loss", two small triangles). With low pass-through the country can come out ahead, but that is a transfer from foreigners, it invites retaliation, and retaliation is not modelled here.
  
  The case for tariffs that economists take seriously is narrower than the political one: a young industry that needs time to learn (the "infant industry" argument, going back to Alexander Hamilton and Friedrich List), national security, or a country large enough to push some of the cost abroad. Each has costs, and the evidence for the first is mixed.

  Exchange rates and the Big Mac
  Trade needs a price between two currencies: the exchange rate. If a given basket of goods cost the same everywhere once converted, the rate would be at purchasing power parity (PPP). The Economist's journalist Pam Woodall invented a light-hearted test in 1986: use a Big Mac as the basket. Divide the local price by the US price and you get the exchange rate at which a burger would cost the same. Compare it with the real rate and you can say a currency looks "cheap" or "dear".
  implied rate = local price ÷ US priceover- or under-valuation = actual dollar price of the burger ÷ US price − 1
  
    Interactive · The Big Mac, July 2026
    
    If the currency strengthens by (negative: weakens)
      
    
    Data: The Economist's Big Mac index, July 2026 edition (published on GitHub by The Economist). Positive slider = the currency strengthens. Raw index, not GDP-adjusted.
  
  Do not take the burger too seriously. It is a thing made with local rent and local wages, so it should be cheaper in poorer countries: a Big Mac in India costs less partly because Indian wages are lower. That is why the Economist also publishes a GDP-adjusted index. The deeper lesson is that over many years exchange rates tend to drift toward PPP, but slowly and noisily, and for a few years they can sit far away.

  The balance of payments: every deficit has a mirror
  The balance of payments is an accounting identity wearing a suit. Every transaction between a country and the rest of the world is recorded twice: what was bought or sold, and how it was paid for. So a country's current account (trade in goods and services, plus income and transfers) and its financial account (buying and selling of assets) must mirror each other. A country that buys more than it sells must pay with borrowing or by selling assets.
  current account = national saving − investmenta deficit means the country invests more than it saves, and foreigners lend the difference
  
    Interactive · Where the gap is financed
    
      A high saver
      A borrower
      Balanced
    
    National saving (% of GDP)
      
    Investment (% of GDP)
      
    
    Stylised countries, for the identity only; the presets are not any real country's figures.
  
  A deficit is not automatically a disease. A fast-growing country that borrows to build factories runs one for good reasons; a country that borrows to consume, or whose lenders lose their nerve, is in a different position. For scale: the US current account deficit in the second quarter of 2026 was about $246 billion, or 3.0% of GDP, per the US Bureau of Economic Analysis (release of 24 September 2026), and it has been in deficit for decades. The mirror image is that foreigners keep buying US assets.

  The impossible trinity and currency crises
  In the early 1960s Marcus Fleming and Robert Mundell (separately) showed that a government cannot have all three of: a fixed exchange rate, free movement of capital and its own interest-rate policy. Any two, never three. It is also called the trilemma, and the reason is mechanical. Suppose you peg to the dollar and cut your interest rate below the dollar's. Savers sell your currency to earn more elsewhere, which pushes it down. To hold the peg you must spend reserves buying your own currency. Reserves are finite.
  
    Gives upKeepsExample
    Own interest rateFixed rate, open capitalEurozone members; Hong Kong, whose dollar has been linked to the US dollar since 1983
    Fixed rateOwn rate, open capitalUS, UK, Japan, Canada (floating)
    Open capitalFixed rate, own rateChina, for long periods (Paul Krugman's example)
  
  
    Interactive · Defend the peg (a toy model)
    A central bank holds its currency fixed. How long do its reserves last if it also cuts rates below the world rate? Outflow per day is assumed proportional to capital mobility × the rate gap. Invented numbers, honest logic.
    Capital mobility
      
    Own rate below the world's (points)
      
    Reserves ($ billion)
      
    
    Real attacks are faster than this model, because the moment traders think a devaluation is coming, selling becomes self-fulfilling.
  
  Three real cases in one table. They are the classic stories, not the only ones.
  
    CaseWhat happened
    UK, 16 Sept 1992Britain's pound was held inside the European exchange-rate mechanism. On "Black Wednesday" the base rate rose from 10% to 12% in the morning, a rise to 15% was promised the same day, yet selling continued, and by evening the UK left the system. The rate returned to 10% the next day. The Treasury later estimated the cost at about £3.3 billion; George Soros made over £1 billion on the other side.
    Thailand, 2 July 1997The baht had been fixed at about 25 to the US dollar. Without the reserves to defend it, Thailand floated it. It fell by more than half, to a low of 56 per dollar in January 1998, and the crisis spread across East Asia. The IMF announced a rescue package of more than $17 billion in August 1997.
    Argentina, 2001-02A one-peso-to-one-dollar link with open capital flows (Krugman's example in 1999) was abandoned in January 2002, in the middle of a deep recession.
  
  The pattern: a peg is a promise to spend reserves to keep a price. It is credible as long as reserves look large next to what could flee, and it fails fast once people stop believing it. Not every float is calm either. The point is not that pegs are bad or floats are good. It is that you have to pick which of the three you give up.

  What a graduate just sees
  
    "Better at everything" does not mean "should make everything": count opportunity cost, not hours.
    Trade raises the size of the pie and rearranges who has the slices. Both parts are true.
    Gains are spread thin across consumers; losses are concentrated on particular workers and places, which is why trade politics is loud.
    A tariff is a tax on your own buyers first. Ask who ends up paying more and who gets paid less.
    A cheap-looking currency may just be a poor country: adjust for income before calling it a bargain.
    Current account = saving − investment. A deficit is a borrowing story, and someone is lending.
    You can have only two of: fixed rate, open capital, own interest rate.
    Pegs end suddenly: when reserves look small beside what could leave, belief becomes the whole game.
  

  Where it shows up next
  
    Development Economics (34/48): whether poor countries should open up or protect, and what the East Asian export growth stories show, are the live arguments of that course.
    Strategy & Competition (34/54): supply chains, offshoring and "which country do I make this in?" are comparative advantage chosen by a firm instead of a government.
    Macroeconomics (34/43): when a central bank raises rates, the currency moves, which moves import prices; the open-economy half of inflation.
    In the world: the price of imported fuel and food, a company's exchange-rate risk, why a remittance is worth more or less, and every news item about a tariff, a currency swing or an IMF programme.
  

  If you want to go deeper
  
    📘 Textbook. Paul Krugman, Maurice Obstfeld and Marc Melitz, International Economics: Theory and Policy (Pearson). The standard undergraduate text, with trade and finance in two halves.
    📘 Textbook. Robert Feenstra and Alan Taylor, International Economics (Worth/Macmillan). The other widely assigned choice, strong on data and evidence.
    📘 Gentler. Dani Rodrik, The Globalization Paradox (2011): the trilemma applied to democracy, nations and open markets. Or Douglas Irwin, Free Trade Under Fire, for the case for trade.
    🌐 Free course. MIT OpenCourseWare 14.581 International Economics I (Spring 2013, Costinot and Donaldson). A graduate course: lecture notes and readings on why countries trade and who wins and loses. Hard, but the reading list is the real literature.
  

  
    The takeaway
    Trade is gainful because countries differ in what they give up to make things, so everyone can be better off, but only on average and only if someone is willing to pay the losers. Currencies add a second lesson: a government can fix the exchange rate, keep money flowing freely, or run its own interest rate, and must give up one. Cheap imports, a strong currency and a trade deficit are three views of the same set of accounts.
    Try this: pick one item you bought this week and find where it was made. Ask who in that country gained from the sale, and who in yours gained or lost from its being made there.
  

  
    Sources & Further Reading
    Sources. Ricardo's cloth-and-wine figures (80/90 and 120/100 labour-years) are the standard textbook version of Principles (1817), chapter 7. Autor, Dorn and Hanson, "The China Syndrome" (NBER w18054; quote: "conservatively, it explains one-quarter of the contemporaneous aggregate decline in US manufacturing employment", 1990-2007) and "The China Shock" (NBER w21906, January 2016). Amiti, Redding and Weinstein, "The Impact of the 2018 Trade War on U.S. Prices and Welfare" (NBER w25672, March 2019). The German comparison is Dauth, Findeisen and Suedekum, Journal of the European Economic Association (2014); this piece relies on the paper's stated focus on "employment effects and net gains". Big Mac data: The Economist's published index (github.com/TheEconomist/big-mac-data), July 2026 edition; origin credited to Pam Woodall, 1986. Balance of payments: US BEA release of 24 September 2026. Trilemma: Fleming (1962) and Mundell (1960-63); Krugman's examples of China, Britain, Canada and Argentina. Black Wednesday and Thai baht details from the standard accounts summarised on Wikipedia (Treasury cost estimate £3.3 billion revised 2005; Soros over £1 billion; baht low of 56 in January 1998; IMF package above $17 billion, 11 August 1997). Every interactive number was computed in node: for example, Ricardo's no-trade consumption is 5.88 (Portugal) and 4.55 (England) of each good, and the break-even prices are 0.889 and 1.2; with the world price at 40 the market gains 400 in total (buyers +1,000, makers -600); a 10-unit tariff with full pass-through costs buyers 550, gives makers 250, raises 200 and loses 100 net.
    
      🌐 MIT OCW 14.581 International Economics I
      🌐 The Economist's Big Mac index data
      📘 Krugman, Obstfeld and Melitz, International Economics
      📘 Rodrik, The Globalization Paradox


Step 12. Strategy & Competition: Why Some Firms Keep Their Profits
------------------------------------------------------------------
Why: How firms compete, when they get too big, and what antitrust does about it.
Most good ideas get copied until the profit is gone. Strategy is the study of what stops that, and competition law is the study of when it has stopped too well.

In Damodaran's January 2026 dataset of listed companies, the whole market kept about 9.7 cents of net profit from each dollar of sales. Software firms kept about 25 cents. Tobacco firms kept about 27. Airlines kept about 2.5 and food retailers about 1.3. Airlines carry passengers across oceans and fly them safely; grocers feed whole cities. They are no less useful than software. They are just in a game where profit is competed away.
  Why? One company makes cola and another makes steel, and both work hard. Strategy asks what lets a firm keep earning above the average, and competition policy asks what society should do when a firm keeps too much.

  
    The course in one box
    
      Who takes itBusiness students (a year-3 core course in MBA and undergraduate business programmes, usually after finance and marketing have been met); economics students as industrial organization; and marketing students, who meet it as marketing strategy. Business schools teach it through cases and frameworks; economics departments teach it through game theory and models. They are the same subject seen from two sides.
      What it is forSeeing why some industries and some firms earn well for decades, and when that profit comes from a better product and when it comes from blocking rivals.
      Comes afterMicroeconomics (34/42) and Game Theory (34/45).
      Leads toIt is the capstone of Business Administration and a late core course in Economics and Marketing. It feeds Pricing (34/61) and Entrepreneurship (34/56).
      What this piece gives youYou will not be able to write a strategy for a firm. You will have a checklist of questions that experienced strategists ask, a feel for why scale and networks create winners, and the economist's view of when a market has too few competitors. Easier first reads: Where Money Actually Comes From (04/02), The Mental Models Top Operators Use (04/07) and Capitalism (31/01). Not legal or investment advice.
    
  

  Why some industries are more profitable: Porter's five forces
  In a 1979 Harvard Business Review article, "How Competitive Forces Shape Strategy", Michael Porter made a plain claim: profit depends less on how clever a firm is than on the structure of the industry it is in. He named five forces that squeeze or protect profit:
  
    Threat of new entrants. If anyone can start, profits are copied away.
    Threat of substitutes. Something different that does the same job caps your price.
    Power of buyers. Big, informed buyers who can switch bargain your price down.
    Power of suppliers. If your inputs come from one source, they take your margin.
    Rivalry. Many equal rivals with high fixed costs fight on price.
  
  A weak force is good for the firm and a strong one is bad. A graduate looks at an industry and scores the five before saying anything about the managers.
  
    Interactive · Score an industry
    Rate each force from 1 (weak, good for profit) to 5 (strong, bad). The presets use my own illustrative ratings for five industries; argue with them. The margin shown is Damodaran's net margin for the matching industry, January 2026.
    
    New entrants
    Substitutes
    Buyers' power
    Suppliers' power
    Rivalry
    
    
  
  The model has critics, and they are worth knowing. Kevin Coyne and Somu Subramaniam argue that it assumes buyers, suppliers and rivals never collude or interact, that value comes only from structure, and that uncertainty is low. Adam Brandenburger and Barry Nalebuff added complementors (the makers of things that make your product more valuable) as a sixth force. Others argue that the model ignores what each firm itself is good at, which the resource-based view puts first. Porter himself updated the framework in a 2008 article. All sides agree that both the industry and the firm matter.

  Competitive advantage: cost, differentiation and the value chain
  Inside an industry, a firm that earns more than its rivals has a competitive advantage. Porter's Competitive Strategy (1980) said there are two basic kinds. Cost advantage: you make the same thing for less, and can either undercut rivals or keep the difference. Differentiation: you make something buyers will pay more for. He also warned of being "stuck in the middle", with neither; economists and managers still argue about how often that is true.
  A neat way to see the two together is to ask what a rival's cost lets you charge. If your product is as good as the rival's, a price war drags the price down to the rival's cost, and your margin is your cost gap. If buyers value yours more, you can charge that premium on top.
  your margin = (rival's cost + your quality premium) − your costcost gap plus premium; a negative number means the rival wins the sale
  
    Interactive · Cost, premium or both
    Two firms sell to the same buyers. A price war pushes the loser's price to its own cost. Buyers value your product $premium more than the rival's. All figures invented.
    Your cost per unit
    Rival's cost per unit
    Premium buyers pay for yours
    
    
  
  Porter's value chain (Competitive Advantage, 1985) tells you where to look: inbound logistics, operations, outbound logistics, marketing and sales, service, supported by technology, procurement, people and infrastructure. An advantage hides in one link or in how links fit. Ikea's cost advantage lives in design and flat-pack logistics, not in cheap wood. A luxury brand's lives in marketing and service. The question is always: which activity do we do better, and can they copy it? A copyable advantage lasts about as long as it takes to copy.

  Economies of scale and the learning curve
  Bigness helps in two plain ways. Fixed costs spread out: a factory costs the same whether it makes a thousand units or a million, so the average cost per unit falls with volume. And experience teaches: the more a firm makes, the better it gets. T. P. Wright noticed in 1936 that the labour per aeroplane fell by a steady percentage each time cumulative output doubled; consultancies later called this the experience curve.
  average cost = fixed cost ÷ volume + unit cost × (learning factor)learning factor = r^d, where r is the cost kept after each doubling of cumulative output and d the number of doublings
  
    Interactive · Scale and learning
    A firm with $F million of fixed cost makes Q million units. Its rival makes half as many and has had one doubling fewer of experience. Invented numbers; the shape is what matters.
    Fixed cost ($ million)
    Volume (million units)
    Cost kept after each doubling
    Doublings of experience (you)
    
    
  
  The catch: scale only helps if demand is large enough to use it. A giant plant sitting half-empty has higher average costs than a small full one. And a learning curve is not automatic; it comes from deliberate work, and rivals can also learn.

  Network effects and platforms
  Some goods get more valuable as more people use them. A telephone is useless alone and valuable with millions. Economists call this a network effect (a demand-side economy of scale): it raises what buyers are willing to pay, rather than lowering the seller's cost. Direct effects come from other users of the same thing (messaging, a phone network). Indirect or two-sided effects link groups: more drivers make a ride app better for riders, and more riders make it better for drivers. A platform is a business built to host such a market, and the economics of pricing two sides at once was set out by Jean-Charles Rochet and Jean Tirole (2003).
  Robert Metcalfe's law says a network's value grows with the square of its users, since n users have about n(n−1)/2 possible connections. Andrew Odlyzko, Bob Briscoe and Benjamin Tilly argued in 2006 ("Metcalfe's Law is Wrong") that later users are worth less than early ones, so growth is nearer n × log n. Studies of Facebook and Tencent have fitted n², and a European internet-usage study found n² for small and n log n for large networks. The data haven't settled it; the main point survives either way.
  
    Interactive · Tipping point
    A toy model. Each person joins if their taste for the product, times its value (a base value plus the network's strength times the share who already use it), exceeds the price. Rounds repeat. Start below the critical mass and it dies; above it, it takes off.
    Network strength
    Price (as a share of top value)
    Starting share of users
    
    
  
  Three cautions. Tipping is not automatic: it needs network benefits to outweigh product differences and costs of switching or using several services at once. Some networks congest (more users can make it worse). And data advantages are often weaker and shorter-lived than classic network effects. The famous cases, such as VHS beating Betamax, are told with different details by different historians.

  Entry, price wars and the economist's view of competition
  Strategy books ask how to win. Industrial organization asks the opposite: how many rivals does it take for a market to behave competitively? The classic model is the French economist Antoine Cournot's (1838): firms choose how much to make, and the price falls as total output rises. With one firm you get the monopoly price. With two, a lower price. As the number of firms grows, price slides toward cost. Prices fall quickly at first and then slowly.
  
    Interactive · How many rivals?
    Demand is P = 100 − Q. Each of N identical firms chooses output; unit cost is c. HHI is the sum of squared market shares. The merge button removes one firm (two become one).
    Number of firms
    Unit cost
    
      Merge two firms
    
    
    
  
  Notice what the model leaves out, and what strategists add. Real firms also enter and exit, and an incumbent may threaten a price war to deter entry. Game theory (34/45) shows that a threat to fight is only believable if fighting is in the incumbent's own interest when the time comes; otherwise entrants call the bluff. Firms avoid price wars when they can: in the Cournot picture profit falls fast as rivals multiply, which is why firms prefer differentiating products to matching prices.

  Market power and antitrust
  Firms that keep profits by being better are doing what the system rewards. Firms that keep them by shutting rivals out are doing something else. Telling them apart is the job of competition law (called antitrust in the United States), and the regimes differ.
  
    RegimeMain rulesNotes
    United StatesSherman Act (1890), Clayton Act (1914), Hart-Scott-Rodino (1976)Sherman §2 bars monopolisation. Mergers are reviewed by the DOJ and FTC.
    European UnionArticles 101 (agreements) and 102 (abuse of dominance); Merger Regulation 139/2004Digital Markets Act applies since 2023 to designated "gatekeepers", fines up to 10% of worldwide turnover.
    ChinaAnti-Monopoly Law, in force since 2008Enforced since 2018 by the State Administration for Market Regulation (SAMR).
    IndiaCompetition Act, 2002Enforced by the Competition Commission of India.
  
  Recent examples, each dated. On 5 August 2024 a US federal judge, Amit Mehta, found that Google had illegally maintained a monopoly in general search; in a remedies decision of 2 September 2025 he declined to break the company up, barred exclusive contracts for search and ordered some data sharing, and both sides appealed. In April 2025 the European Commission imposed the first fines under the Digital Markets Act: €500 million on Apple and €200 million on Meta. In April 2021 China's regulator fined Alibaba about $2.8 billion, around 12% of its 2020 net profit, for anticompetitive practices.
  The tools differ in what they measure. A common first screen is the HHI (the widget above). The US 2010 guidelines treated a market above 2,500 as highly concentrated; the 2023 revision tightened the threshold, so check the current text. The deeper argument is about the goal. The consumer welfare standard, associated with Robert Bork, asks whether prices rise or output falls for buyers. Critics (sometimes called "neo-Brandeisian") say this is too narrow and misses power over workers, suppliers and politics. Defenders say it is the only standard that can be measured and tested. Most jurisdictions now treat consumer welfare as one factor among several.

  What a graduate just sees
  
    Industry structure explains a lot of profit before you look at the managers: score the five forces.
    An advantage is only as good as it is hard to copy.
    Your margin = rival's cost + your premium − your cost: an advantage is a gap.
    Average cost falls with volume, but only if demand fills the factory.
    Network effects create winners by tipping; critical mass is the whole game, and expectations decide it.
    In a market with few rivals, prices drift up; each extra rival helps, with diminishing force.
    A merger changes the concentration number, and regulators look at both the level and the change.
    "Dominant because better" and "dominant because blocked everyone else" can look the same from outside, and the law tries to tell them apart.
  

  Where it shows up next
  
    Pricing (34/61): once you know your cost and your rival's cost, pricing is choosing where in that gap to sit.
    Marketing Management (34/57): positioning is differentiation seen from the customer's side.
    Entrepreneurship (34/56): a new venture's first question is which force it can beat and for how long.
    In the world: why airlines and grocers earn thin margins, why a "winner takes most" app market forms, what a merger review asks, and every news story about a company being sued by a regulator.
  

  If you want to go deeper
  
    📘 Textbook. Robert Grant, Contemporary Strategy Analysis (Wiley). The standard strategy text for business students, with industry analysis, resources and cases.
    📘 Textbook. David Besanko, David Dranove, Mark Shanley and Scott Schaefer, Economics of Strategy (Wiley). The economist's version, with game theory throughout. For the pure theory, Jean Tirole's The Theory of Industrial Organization (MIT Press, 1988).
    📘 Gentler. Michael Porter, Competitive Strategy (1980), for the originals; Carl Shapiro and Hal Varian, Information Rules (1999), for networks and standards; Tim Wu, The Curse of Bigness (2018), for the antitrust story and the argument against consumer welfare.
    🌐 Free course. MIT OpenCourseWare 15.902 Strategic Management I (Fall 2006, Hax and Sahani), a graduate Sloan course on analytical methods and strategic practice.
  

  
    The takeaway
    Profit lasts where something stops rivals: high entry barriers, scale a rival cannot match, a network that tips, or a product buyers will not swap. Where nothing stops them, profit is competed down to cost. The economist adds a second question: when a firm's profit comes from blocking rivals rather than serving buyers, does society step in, and by what test?
    Try this: pick a business you use (a bank, a streaming service, a shop). Score its five forces from 1 to 5, then name the one thing that stops a rival from copying it. If you cannot name one, expect thin margins.
  

  
    Sources & Further Reading
    Sources. Margins: Aswath Damodaran, industry margins dataset, January 2026 (net margins: all firms 9.74%, software 25.49%, tobacco 26.65%, soft drinks 13.40%, air transport 2.51%, steel 1.93%, grocery and food retail 1.32%, autos and trucks 1.29%); the five-force ratings in the widget are the author's. Five forces: Porter, "How Competitive Forces Shape Strategy", Harvard Business Review (1979) and "The Five Competitive Forces That Shape Strategy" (2008); criticisms (Coyne and Subramaniam; Brandenburger and Nalebuff's complementors; the resource-based view) as summarised on Wikipedia. Generic strategies (1980) and the value chain (1985) are standard textbook attributions to Porter. Wright (1936) is the standard origin of the learning curve. Network effects: definitions and tipping caveats as summarised on Wikipedia; Metcalfe's law and its critics (Briscoe, Odlyzko and Tilly, IEEE Spectrum, 2006; Madureira et al. 2013; Zhang, Liu and Xu 2015) likewise; Rochet and Tirole (2003) on two-sided markets. Antitrust: statutes as listed on Wikipedia's competition-law article; United States v. Google (Mehta, 5 August 2024 liability ruling; 2 September 2025 remedies); Digital Markets Act (in force 1 November 2022, gatekeepers designated September 2023, first fines April 2025: Apple €500 million, Meta €200 million); Alibaba fine, SAMR, 9 April 2021 (as reported, $2.8 billion, about 12% of 2020 net profit); US HHI bands from the 2010 Horizontal Merger Guidelines (1,500 and 2,500). Every interactive number was computed in node: for example, with three firms and unit cost 20 the Cournot price is 40, each firm makes 20 and earns 400, HHI 3,333; merging two raises the price to 46.7; the tipping model with strength 1.5 and price 0.3 has a critical mass of 4.2% and a stable share of 79.1%.
    
      🌐 MIT OCW 15.902 Strategic Management I
      🌐 Damodaran: margins by industry
      📘 Besanko et al., Economics of Strategy
      📘 Tim Wu, The Curse of Bigness


Step 13. Corporate Finance: The Price of Time and Risk
------------------------------------------------------
Why: How markets price time and risk.
A pound today is worth more than a pound next year, and a risky pound is worth less than a safe one. Corporate finance is the craft of putting a number on both, so that a company can decide which projects are worth their money.

Someone offers you $1,000 today or $1,100 in a year. Almost nobody needs a course to say "it depends": on whether you trust them, on what else you could do with the money, on whether you need it now. Those three words, trust, alternatives and need, are the whole subject. Corporate finance is what happens when you ask the question about a factory instead of a loan, with a hundred million at stake and a board waiting for the answer.
  Every company faces the same three decisions. Which projects should we spend on? Where does the money come from, borrowed or from owners? And what is the whole business worth? A graduate has one tool for all three: a way of turning cash that arrives at different times, with different risks, into a single number today.

  
    The course in one box
    
      Who takes itBusiness, economics and engineering students, usually in year 2 or 3; every MBA takes it in the first year, and it sits in Wharton's undergraduate business core and in Harvard Business School's Required Curriculum. Finance majors then split it into investments, derivatives and banking.
      What it is forDeciding whether a project, a company or a funding plan is worth more than it costs, once time and risk are priced in.
      Comes afterAccounting (34/50) for the cash flows, and Probability & Statistics (34/05) for risk.
      Leads toEntrepreneurship (34/56). Civil and other engineers meet it as "engineering economics" when they ask whether a bridge is worth building.
      What this piece gives youYou will not be able to value a real company. You will be able to compute a net present value, see why a headline return can mislead, and know what a discount rate is made of. Easier first reads in the library: Compound Interest Will Blow Your Mind (03/03) and Investing Without Gambling (03/10). This piece goes further. It describes how finance works and what economists argue about; it is not investment advice.
    
  

  Money has a time stamp: compounding and present value
  If a bank pays 6% a year, $1,000 today becomes $1,060 next year and $1,123.60 the year after, because the interest earns interest. That is compounding. Run it backwards and you get its twin, present value: what a future sum is worth now. At 6%, a promise of $1,000 in 20 years is worth only about $312 today, because $312 invested at 6% would grow to $1,000 by then.
  FV = PV × (1 + r)n   ·   PV = FV ÷ (1 + r)nr = interest rate per period, n = number of periods. The rate that turns future money into today's money is called the discount rate.
  
    Interactive · Grow it, then discount it
    Amount
      
    Interest or discount rate
      
    Years
      
    
  
  At the default, $1,000 grows to $3,207 in 20 years, and $1,000 due in 20 years is worth $312 now. The rule of 72 gives a shortcut: money doubles in roughly 72 divided by the percentage rate, so about 12 years at 6% (the exact answer is 11.9). The same arithmetic is why waiting costs: a rate is the price of time.

  Net present value: the one rule for saying yes
  A project is a stream of cash: an outlay now, receipts later. To judge it, discount each receipt back to today and add them up. The sum, minus the outlay, is the net present value (NPV). If NPV is positive, the project earns more than the discount rate demands, and doing it makes the owners richer by exactly that amount. If it is negative, they would do better lending the money at the discount rate.
  NPV = −Outlay + CF1/(1+r) + CF2/(1+r)2 + … + CFn/(1+r)naccept when NPV > 0. The discount rate is your opportunity cost for projects of this risk.
  Worked: spend $1,000 now for $300 a year over five years. At an 8% discount rate the five receipts are worth $1,198 today, so NPV is +$198. At 15% the receipts are worth $1,006, and NPV is nearly zero. The rate at which NPV hits zero is the project's internal rate of return (IRR), here 15.2%. Slide the rate and watch the line cross zero.
  
    Interactive · NPV against the discount rate
    Outlay today
      
    Cash per year
      
    Years
      
    Discount rate
      
    
    
  

  When IRR misleads
  IRR is popular because it is one percentage, easy to compare with a hurdle rate. It is also the part of finance where the textbooks warn you most. Two classic failures.
  Several IRRs. If the cash flow changes sign more than once, for example a mine that needs a big cleanup bill at the end, the NPV curve can cross zero twice. A project with flows of −100, +230 and −132 (in millions) has IRRs of both 10% and 20%, and it is worth doing only when the discount rate lies between them. A single IRR number hides that.
  Scale. Project A turns 100 into 150 (IRR 50%). Project B turns 1,000 into 1,300 (IRR 30%). If you can do only one and the cost of capital is 10%, A adds 36 and B adds 182. Higher IRR, much lower value. The two NPV curves cross at about 27.8%, and below that rate B wins. The rule graduates keep: when IRR and NPV disagree, trust NPV, because it measures money, not percent.
  
    Interactive · Two traps
    
      Cleanup bill at the end
      Small and fast vs big and slow
    
    Discount rate
      
    
    
    Illustrative cash flows (millions), invented to show the traps.
  

  Risk and return, and why diversification is nearly free
  So far the future was certain. It is not. A share can fall 40% in a year, and investors want to be paid for taking that chance. The central observation of modern finance, made by Harry Markowitz in 1952, is that you are paid for the risk you cannot avoid, and that you can avoid a lot of it for nothing. Hold many assets that do not move in lockstep and their ups and downs partly cancel.
  Risk here is measured as volatility: the standard deviation of returns. For a portfolio of N assets, each with volatility σ and an average pairwise correlation ρ, the portfolio's volatility is:
  σp = σ × √( 1/N + (1 − 1/N) × ρ )diversification removes the 1/N part; the ρ part, risk shared by all assets, is what is left
  Take stocks with 30% volatility each and a correlation of 0.3. One stock: 30%. Ten stocks: 18.2%. Fifty: 16.8%. The floor is 30% × √0.3 = 16.4%, no matter how many you hold. That floor is market risk, which cannot be diversified away; the rest is specific risk, which can.
  
    Interactive · How many stocks is enough?
    Each stock's volatility
      
    Correlation between stocks
      
    Number of stocks held
      
    
    
    Assumes equal volatility and equal pairwise correlation, a simplification. Real correlations vary and rise in crises.
  

  CAPM: the idea, and its critics
  If investors diversify, the only risk worth a reward is the market risk left over. The Capital Asset Pricing Model, developed in the early 1960s by Jack Treynor, William Sharpe, John Lintner and Jan Mossin (Sharpe's version appeared in 1964), turns that into a price. A stock's beta measures how strongly it moves with the market: beta 1 moves with it, 0.5 half as much, 1.5 one and a half times. The model says the required return is the safe rate plus beta times the market's extra reward:
  Required return = rf + β × (market return − rf)the market return minus rf is called the equity risk premium
  
    Interactive · The security market line
    Risk-free rate
      
    Equity risk premium
      
    Beta
      
    
    Inputs are illustrative, not forecasts. Estimates of the premium range widely.
  
  With a 3% safe rate, a 5% premium and a beta of 1.2, the required return is 9%. That number is the discount rate you use in NPV. The model is elegant, which is why it is taught everywhere, and why it is argued about. Critics: Fama and French (1992) found that beta alone did a poor job of explaining average returns across US stocks, while company size and the ratio of book value to price did better, and Fama and French later added those as extra factors. Richard Roll (1977) argued the model cannot truly be tested, since the "market" should hold every asset, including property and human skills. Many practitioners use CAPM anyway as a rough, transparent starting point. Treat the number as an estimate with a wide error bar.

  Cost of capital and valuation: discounted cash flow
  A company's cost of capital is the return its funders demand: lenders want interest, owners want a CAPM-style return. Blend the two, weighted by how much each provides, and adjust the debt for tax, and you get the weighted average cost of capital (WACC). It is the discount rate for the firm's cash flows.
  A discounted cash flow (DCF) valuation then does what NPV did for projects, but for the whole firm. Forecast the free cash flow for some years, add a terminal value for everything after, and discount it all. With this firm's cash of $100 million, 8% growth for five years, 2% afterwards and a 9% WACC, the business is worth about $1.88 billion, and 74% of that comes from the terminal value. That is the first lesson of DCF: most of the answer is the distant future, so the answer is only as good as the growth and discount-rate assumptions. Move the terminal growth up to 4% and the value jumps by almost a third.
  
    Interactive · Value a company
    Growth for the next 5 years
      
    Growth forever after
      
    Discount rate (WACC)
      
    
    A company with $100 million of free cash flow now. If growth forever would reach the discount rate, the tool caps it one point below.
  
  The other common method uses multiples: look at what similar companies sell for relative to a measure like earnings or cash flow, and apply the same ratio. The tool above would call this firm's value "18.8 times cash flow". Multiples are quick and market-based, but they only move with the market: if the whole sector is overpriced, so is yours. Careful analysts use both and ask why they disagree.

  Capital structure: Modigliani and Miller in words
  Should a firm fund itself with debt or equity? In 1958 Franco Modigliani and Merton Miller proved something startling: in a world with no taxes, no bankruptcy costs and no information gaps, it does not matter. The firm's value comes from what its assets earn, not from how the claims on them are sliced. An investor could borrow personally to copy a leveraged firm, so the market will not pay extra for the firm doing it. More debt makes the equity riskier, so shareholders demand a higher return that exactly cancels the cheapness of debt. They received Nobel prizes for this (Modigliani 1985, Miller 1990).
  The theorem is a benchmark: it tells you where to look for real effects. The first is tax. In most countries interest is deductible and dividends are not, so debt carries a tax shield and a levered firm's WACC falls. The second is distress: more debt raises the chance of bankruptcy, which is costly. The textbook "trade-off theory" balances the two; pecking order theorists argue managers instead prefer internal cash, then debt, then new shares, because of information gaps. Both are live in the literature.
  
    Interactive · Debt, tax and the cost of capital
    Debt as a share of firm value
      
    Corporate tax rate
      
    
    Assumes an all-equity return of 10% and debt at 5%. Distress costs are not modelled, which is why the tax benefit looks endless here. Real firms hit a limit.
  
  With no tax, the WACC stays at 10% at any debt level while the cost of equity climbs (13.3% at 40% debt). With a 25% tax rate, 40% debt cuts WACC to 9%. Pushed to 90% the tool says 7.75%, which no lender would permit: that is the limit the distress cost supplies.

  Efficient markets: the debate that never ended
  If prices already reflect what everyone knows, nobody can reliably beat the market, and an NPV calculation is just reading the price. Eugene Fama's 1970 review gave this efficient market hypothesis its structure: weak (past prices carry no news), semi-strong (public information is already in the price) and strong (even private information is). The evidence is mixed in a way that keeps both camps in business.
  
    For: most actively managed funds trail a plain index over long periods, which the S&P SPIVA scorecards have reported year after year across many markets.
    Against: Robert Shiller showed in 1981 that stock prices move far more than later dividends justify, and bubbles and crashes seem hard to explain with rational pricing. Behavioural economists, such as those in Behavioral Economics (34/46), build on this.
    A middle path: Sanford Grossman and Joseph Stiglitz argued in 1980 that markets cannot be perfectly efficient, because then nobody would pay to gather the information that makes them efficient.
  
  Fama, Shiller and Lars Peter Hansen shared the 2013 Nobel prize in economics: a prize for one side's theory, the other side's critique and the statistical methods both used. Luck matters more than people think. Try it: give a thousand "fund managers" a coin each and see how many beat the market ten years in a row by chance alone.
  
    Interactive · Skill or coin flips?
    Managers
      
    Years in a row
      
    Flip the coins
    
    Each manager has a 50% chance of beating the market each year, independent of the others. The simulation uses your browser's random numbers.
  
  For 1,000 managers and ten years the expected number with a perfect record is about one (1,000 ÷ 1,024), and there is a 62% chance of at least one. That one looks like a genius. Real streaks are not purely luck, but the arithmetic is why careful readers ask how many managers started.

  What a graduate just sees
  
    A sum of money has a date on it. Never add cash flows from different years without discounting.
    A discount rate is an opportunity cost: what else the money could earn at the same risk.
    NPV measures value created in dollars; IRR is only a percentage and can mislead, especially with scale or sign changes.
    You are paid for risk you cannot diversify away, not for risk you could have removed for free.
    A valuation is mostly about the distant future, so ask which assumption carries it.
    Debt is cheaper because it is safer for lenders, which makes equity riskier. The free lunch is in the tax code, not in the maths.
    Beating the market is hard; many apparent winners are lucky survivors.
    Every number has a model behind it. Know which, and what it assumes.
  

  Where it shows up next
  
    Entrepreneurship (34/56): equity, dilution and the valuation of a company that does not yet earn.
    Economics and Macroeconomics (34/42, 34/43): the interest rate is set by savers, firms and central banks; every discount rate in this piece sits on top of it.
    Behavioral Economics (34/46) and Game Theory (34/45): where the efficient-market story strains, and how markets with strategic players behave.
    In the world: a mortgage, a pension, a merger announcement, a startup funding round, a new power plant, a government deciding whether a railway pays back.
  

  If you want to go deeper
  
    📘 Textbook. Richard Brealey, Stewart Myers, Franklin Allen and Alex Edmans, Principles of Corporate Finance (McGraw Hill, 14th edition, 2022). Strong on intuition and real cases.
    📘 Textbook. Jonathan Berk and Peter DeMarzo, Corporate Finance (Pearson, current edition). More systematic, built around no-arbitrage reasoning.
    📘 Gentler. Peter Bernstein, Against the Gods: The Remarkable Story of Risk (Wiley, 1996) for the history of risk; Burton Malkiel, A Random Walk Down Wall Street (Norton) for the efficient-markets argument in plain language.
    🌐 Free course. MIT OpenCourseWare 15.401 Finance Theory I (Fall 2008, Andrew Lo): lectures and notes on present value, risk, portfolio theory and valuation.
  

  
    The takeaway
    Corporate finance turns cash that arrives at different times and different risks into a single number today. Discount at a rate that reflects what the money could earn elsewhere at the same risk; trust NPV over IRR; remember that diversification removes risk for free and the rest is what you are paid for. Every valuation is a model, and its biggest input is usually a guess about the far future.
    Try this: take any offer with delayed money (a "0% for 12 months" deal, a pension quote, a subscription that saves 20% if you pay a year ahead). Pick a discount rate you believe, and compute the present value of both sides.
  

  
    Sources & Further Reading
    Sources. Time value, NPV, IRR and the diversification formula are standard textbook finance (Brealey et al.; Berk & DeMarzo). Markowitz, "Portfolio Selection", Journal of Finance (1952); Sharpe, "Capital Asset Prices", Journal of Finance (1964), with Treynor, Lintner and Mossin independently; Roll, "A Critique of the Asset Pricing Theory's Tests" (1977); Fama and French, "The Cross-Section of Expected Stock Returns", Journal of Finance (1992); Modigliani and Miller, "The Cost of Capital, Corporation Finance and the Theory of Investment", American Economic Review 48(3) (1958); Fama, "Efficient Capital Markets: A Review of Theory and Empirical Work" (1970); Shiller (1981) on excess volatility; Grossman and Stiglitz (1980). Nobel prizes: Markowitz, Miller and Sharpe 1990; Modigliani 1985; Fama, Hansen and Shiller 2013. The SPIVA claim refers to S&P Dow Jones Indices' published scorecards; the share varies by year and market, so no percentage is quoted. Interactive numbers were computed in node: $1,000 at 6% for 20 years = $3,207.14, present value $311.80, doubling 11.90 years; NPV of 5 × $300 on $1,000 at 8% = $197.81, IRR 15.24%; the −100/+230/−132 flows have IRRs of exactly 10% and 20%; A versus B at 10% = $36.36 against $181.82, crossover 27.8%; diversification 30.0%, 18.2%, 16.8% and floor 16.4%; CAPM 3% + 1.2 × 5% = 9%; DCF enterprise value $1,877.9 million with terminal 74.1%, 2,472.7 at 4% terminal growth; WACC 9.0% and cost of equity 12.5% at 40% debt and 25% tax; 1,000/1,024 = 0.98 managers, 62.4% chance of at least one.
    
      🌐 MIT OCW 15.401 Finance Theory I
      📘 Brealey, Myers, Allen & Edmans, Principles of Corporate Finance
      📘 Bernstein, Against the Gods
      📘 Malkiel, A Random Walk Down Wall Street


Step 14. Development Economics: Why Some Countries Are Rich, and What Has Actually Helped
-----------------------------------------------------------------------------------------
Why: Why some countries are rich and others poor, and what has actually helped.
In 1960 Ghana looked richer than South Korea. Sixty years on, Korea's income per person is about ten times Ghana's. The subject asks why, and which fixes have been tested rather than just believed.

On a world map of average income, two peninsulas stand out. South Korea and North Korea share a language, a climate, a history up to 1945 and most of their genes. Their incomes today are wildly different. A single border drawn across one place did that, which is hard to explain by weather or by who the people are.
  Or take Ghana and South Korea. By the Maddison Project's estimates (2023 edition, in 2011 international dollars), Ghana's income per person in 1960 was about $2,200 and Korea's about $1,500. By 2022 Korea was at about $41,300 and Ghana at about $4,200. Nobody sat down and chose that. Development economics is the attempt to find out what produced it, and what a government, an aid agency or a charity can actually do about it.

  
    The course in one box
    
      Who takes itEconomics majors as a year-4 field course; students of international development, public policy and global health; many practitioners in aid agencies and ministries. It is taught in economics departments and in policy schools, and it is where economics most often meets field data. Some universities teach it as an undergraduate course (MIT's 14.73, The Challenge of World Poverty, is one); others only at graduate level.
      What it is forUnderstanding why incomes differ so much between countries, how poverty is counted, and which interventions have evidence behind them.
      Comes afterMacroeconomics (34/43) and Econometrics (34/44).
      Leads toIt is the frontier course of the economics major. It draws on International Trade & Finance (34/47) and Public Economics (34/49) and ends in practice: working in a ministry, a development bank or a research group.
      What this piece gives youYou will not be able to design a development programme. You will know the three big explanations of rich versus poor and why economists have not settled between them; how poverty is counted; what a randomised trial can and cannot tell you; and what the evidence on aid, cash transfers and microfinance actually says. Easier first reads: The Mixed Economy (31/05) and Institutions: The Invisible Rules That Shape Us (30/05). This one adds the data.
    
  

  Why some countries are rich: geography, institutions, human capital
  Three families of explanation compete, and each has a serious case.
  Geography. Where a country sits shapes its costs. John Gallup, Jeffrey Sachs and Andrew Mellinger argued in 1998 that geography affects growth directly, through transport costs (landlocked countries far from the coast pay more to trade), disease burdens (the tropics carry more) and farm productivity. Jared Diamond's Guns, Germs, and Steel (1997) made a broader version of the case about continents and crops.
  Institutions. The rules of the game matter more. Daron Acemoglu, Simon Johnson and James Robinson argued in "The Colonial Origins of Comparative Development" (2000, published 2001) that where Europeans could not survive diseases they built extractive institutions, which lasted; where they could settle in numbers they built ones that protected property. Mortality among early settlers, they said, predicts institutions today, and institutions predict income. The three won the 2024 Nobel prize in economics for their work on how institutions shape prosperity. Korea against North Korea is the standard picture.
  Human capital. People matter more. Edward Glaeser, Rafael La Porta, Florencio Lopez-de-Silanes and Andrei Shleifer (2004) argued that schooling is the deeper driver and that institutions follow growth; poor countries, they said, have often escaped poverty through sensible policies, sometimes under dictators, and improved their political institutions later.
  
    SchoolIts strongest caseThe standard objection
    GeographyCoast access and disease burden predict incomeKorea and North Korea share geography; places change with technology
    InstitutionsSettler mortality predicts institutions, and institutions predict incomeCritics question the mortality data, and ask whether it works only through institutions
    Human capitalSchooling predicts growth and precedes institutional changeRich countries can afford more schooling, so the causation can run the other way
  
  Most economists weigh all three. The live argument is over which is deepest: what causes the others. That has not been settled, and each side's best evidence is read differently by the others.

  Growth miracles and stalls
  Whichever school you favour, the arithmetic is shared. Income growing at g per cent a year doubles in about 70 ÷ g years. At 2% a country doubles in 35 years; at 7%, in ten. Small gaps in growth, sustained, produce giant gaps in income: that is the whole reason the subject exists.
  years to double ≈ 70 ÷ growth rate (%)the "rule of 70": exactly, ln 2 ÷ ln(1 + g)
  
    Interactive · The long race
    Income per person in constant international dollars (2011 prices). Pick a country and a window. The US is the yardstick. Source: Maddison Project Database 2023, as published by Our World in Data. Figures before about 1990 for poor countries are rough estimates.
    
    From
      
    To
      
    
  
  Read the table the way a graduate does. Korea (5.4% a year), Taiwan (5.3%), Botswana (5.6%) and China (4.8%) grew 18 to 29 times richer from 1960 to 2022. Vietnam and India grew about six times richer. Ghana (1.1% a year) grew about 1.9 times; Niger shrank slightly. Argentina managed 1.2% over the period, and Brazil and Mexico were flat through the 1980s, the "lost decade". Venezuela's income per person fell from about $17,100 in 2010 to about $5,300 in 2022, which is a collapse, not a stall. Even the United States, at 1.9% a year, became 3.2 times richer.
  One caution: the "middle-income trap" (countries growing quickly and then stalling) is a popular phrase, and economists dispute how real it is as a separate pattern. The data show stalls, and the explanations fight.

  Measuring poverty: where the line is drawn
  To say how many people are poor, you need a line. The World Bank's international poverty line began in 1990 at about a dollar a day and has been re-based as price data improved: $1.90 (2011 prices), $2.15 (2017 prices) and, from June 2025, $3.00 a day in 2021 prices, with higher lines of $4.20 and $8.30 for lower-middle and upper-middle-income countries. The dollars are adjusted for what money buys locally (purchasing power parity, as in International Trade & Finance (34/47)), so a "dollar a day" is not a dollar's worth of food at the exchange rate.
  On the $3.00 line, the World Bank's September 2026 update says about 811 million people live in extreme poverty, roughly one in ten, down from about 2.3 billion in 1990. Progress has slowed in the past decade, and extreme poverty is increasingly concentrated in Sub-Saharan Africa and in conflict-affected, fragile countries. Critics argue the line is far too low: Peter Edward of Newcastle University, for example, argues for a figure around $7.40 a day. Where you put the line changes the story.
  
    Interactive · Move the line
    An illustrative income distribution (not a real country): incomes are log-normal, a shape real distributions roughly follow. Change the line, the middle income, or inequality, and see who counts as poor.
    
      $2.15
      $3.00
      $4.20
      $8.30
    
    Poverty line ($/day)
      
    Median income ($/day)
      
    Inequality (spread of incomes)
      
    
    
  
  Two things the slider teaches. Raising the median income lowers poverty a lot, which is why growth matters. But the same median with more inequality puts more people below a low line. Both levers move the count; neither is the whole story.

  The randomista turn: Banerjee, Duflo and Kremer
  For decades the big questions were answered with cross-country regressions, and the answers were fragile: there are too few countries and too many things that differ between them. In the 1990s and 2000s a group of economists took a different route: run experiments. Pick villages, schools or households at random for a programme, leave the rest as a comparison, and measure the difference. Abhijit Banerjee, Esther Duflo and Michael Kremer shared the 2019 Nobel prize in economics for this experimental approach to alleviating global poverty. Their sceptical friends called the movement the "randomistas".
  The logic is plain: random assignment makes the two groups alike on average, so a difference afterwards is caused by the programme. What stops it being easy is noise. A small trial might miss a real effect, and a lucky one might exaggerate it. That is what statistical power is, and you can feel it.
  
    Interactive · Run a trial
    A programme has a true effect (in standard deviations of the outcome). You enrol n people in each group. Each press runs 200 imaginary trials with random noise and counts how many reach "p < 0.05".
    True effect (standard deviations)
      
    People per group
      
    Run 200 trials
    
    
    Estimate = difference in group means, each person's outcome with standard deviation 1. Standard error = √(2/n). Gold bars are the trials that "found" an effect; dashed line is the truth.
  
  Try 0.2 with 100 people per group: only about 29% of trials detect a real, modest effect, and the ones that do overstate it, averaging about 0.37. A "significant" result from a small study is a biased reading of the truth. That is why economists ask for large samples and replication.
  Now guess what trials found. Each of these four ran in Kenya or across several countries, and each tested something many people expected to work.
  
    Interactive · Predict, then see
    
    
    
      I think so
      I doubt it
    
    
  

  The critics: what Deaton said
  Angus Deaton, who won the Nobel prize in economics himself in 2015 for work on consumption and poverty, was an early and sharp critic. In a 2009 paper with a title that is itself an argument, "Instruments of development: Randomization in the tropics, and the search for the elusive keys to economic development", he argued that RCTs and similar methods often misread what makes assignment random, that average effects hide variation, and that "experiments have no special ability to produce more credible knowledge" than other methods. His conclusion was that project evaluation by RCT is unlikely to advance understanding of economic development, and he welcomed a move toward testing theoretical mechanisms.
  The randomistas' reply: small, answerable questions add up. You may not learn why Korea grew, but you can learn whether a net should be free or priced, and the answer is usable. The result is a split in the field about what counts as knowledge:
  
    QuestionExperimentalistsDeaton's side
    What can we know?Specific effects of specific programmesMechanisms and theory matter more
    Does it generalise?Replicate in many placesContext and heterogeneity can undo it
    Does it answer big questions?Not directly; it does answer practical onesThen the big questions go unanswered
  
  Neither side has won; most working development economists use both kinds of evidence.

  Aid, cash and microfinance
  Aid. In 1970 the UN General Assembly set a target for rich countries to give 0.7% of their national income as aid. The argument about whether aid works is older than the target. Jeffrey Sachs argues that aid, if it tackles disease, low farm yields and high transport costs, can break a poverty trap. William Easterly argues that top-down aid distorts incentives and does little. Dambisa Moyo argues that aid breeds corruption and dependency and that African countries should lean on private capital. Statistical studies of aid and growth give mixed results; early ones often found little or negative effects and later ones more positive but inconclusive.
  Cash. An alternative to projects is simply giving poor households money. In a large Kenyan experiment, Dennis Egger, Johannes Haushofer, Edward Miguel, Paul Niehaus and Michael Walker (published 2022) found a local transfer multiplier of about 2.4, minimal price inflation and "large positive spillovers on non-recipient households and firms". Cash did not simply raise local prices and hurt the neighbours.
  Microfinance. Small loans to the poor, associated with Muhammad Yunus and Grameen Bank, were hailed as a breakthrough. Six randomised evaluations in six countries (Banerjee, Karlan and Zinman, 2015) found a "consistent pattern of modestly positive, but not transformative, effects". Microcredit is a useful tool and not a miracle.
  The sober picture: economists have good evidence on some small things (free nets, cash), mixed evidence on big things (aid), and humility about the biggest question of all.

  What a graduate just sees
  
    Differences in income are enormous and mostly about growth rates sustained over decades: 70 ÷ g gives the doubling time.
    Geography, institutions and human capital are all real; the dispute is over which is deepest.
    A "poverty line" is a choice, in dollars adjusted for local prices, and moving it changes the count.
    Growth lifts the most people out of poverty; but the same average can hide very different inequality.
    A random assignment gives an honest comparison, but a small trial is noisy and "significant" results exaggerate.
    An RCT answers "did this work here?"; it does not automatically answer "will it work there?"
    What people expect to work often does not: charging a small fee for nets, free textbooks, big microcredit effects.
    Cash transfers have held up better than many feared, with local spillovers rather than inflation.
  

  Where it shows up next
  
    Econometrics (34/44): the credibility revolution (natural experiments, instruments, trials) is the toolkit this course is built on.
    International Trade & Finance (34/47): whether a poor country should open its borders or protect an infant industry is one of the oldest fights in development.
    Public Economics (34/49): cash transfers, taxation and public goods in low-income countries are the same questions as in rich ones, with fewer resources.
    In the world: a job in a ministry, a development bank, a research group such as J-PAL (co-founded by Banerjee and Duflo), or an NGO choosing what to fund; and every news story about aid cuts, poverty counts or a new "miracle".
  

  If you want to go deeper
  
    📘 Textbook. Debraj Ray, Development Economics (Princeton University Press, 1998). The rigorous standard, strong on theory.
    📘 Textbook. Michael Todaro and Stephen Smith, Economic Development (Pearson). A widely assigned, broader survey with country cases.
    📘 Gentler. Abhijit Banerjee and Esther Duflo, Poor Economics (2011): the randomista view, told through the poor people's choices. Pair it with Angus Deaton, The Great Escape (2013), for the sceptical view.
    🌐 Free course. MIT OpenCourseWare 14.73 The Challenge of World Poverty (Spring 2011, Duflo and Banerjee), an undergraduate course covering growth, famine, schooling, microfinance and aid. Also Marginal Revolution University's Development Economics video course (Tyler Cowen and Alex Tabarrok).
  

  
    The takeaway
    Rich and poor countries differ because small differences in growth compound for decades, and the three great explanations (geography, institutions, people) each hold part of the answer. Poverty is counted against a line you have to choose. Randomised trials gave the field honest small answers and sharpened its doubts about big ones. The honest summary is a short list of things that work, a longer list of things that don't as hoped, and a lot of humility.
    Try this: pick a country you care about and find its GDP per person in 1990 and now. Work out the annual growth rate and the doubling time, then ask what would have had to be different for it to grow one point faster.
  

  
    Sources & Further Reading
    Sources. Income per person: Bolt and van Zanden, Maddison Project Database 2023 (constant 2011 international dollars), via Our World in Data; poor-country figures before about 1990 are estimates. Geography: Gallup, Sachs and Mellinger, NBER w6849 (1998). Institutions: Acemoglu, Johnson and Robinson, NBER w7771 (2000) and American Economic Review (2001); 2024 Nobel prize to Acemoglu, Johnson and Robinson. Human capital: Glaeser, La Porta, Lopez-de-Silanes and Shleifer, NBER w10568 (2004). Poverty line history ($1 in 1990; $1.90, $2.15, $3.00; $4.20 and $8.30 lines) and the World Bank's September 2026 update (about 811 million below $3.00 a day, one in ten; about 2.3 billion in 1990): World Bank, Poverty overview page; line history as summarised on Wikipedia. Peter Edward's argument as reported there. 2019 Nobel prize: Banerjee, Duflo, Kremer. Deaton: NBER w14690 (January 2009; quote from the abstract). Aid target: UN General Assembly, 1970. Sachs, Easterly and Moyo positions as summarised on Wikipedia's foreign-aid article. Cash transfers: Egger et al., NBER w26600, Econometrica (2022). Microcredit: Banerjee, Karlan and Zinman, American Economic Journal: Applied Economics (2015). Bednets: Cohen and Dupas, NBER w14406, Quarterly Journal of Economics (2010). Textbooks: Glewwe, Kremer and Moulin, NBER w13300. Every interactive number was computed in node: the growth table (Korea 26.7 times and 5.44% a year, Ghana 1.9 times and 1.07%, Venezuela fell to 0.43 of 1960), the power of a 0.2-standard-deviation effect at 100 per group (29.3%), the average significant estimate (about 0.37).
    
      🌐 MIT OCW 14.73 The Challenge of World Poverty
      🌐 Marginal Revolution University: Development Economics
      📘 Banerjee and Duflo, Poor Economics
      📘 Deaton, The Great Escape


Step 15. Machine Learning: Programs That Write Their Own Rules
--------------------------------------------------------------
Why: Economists now predict and measure with machine learning; knowing its limits matters.
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