Log inStart free

C04

Data, Regression & Decisions

Turn data into decisions: decision trees, testing, regression, forecasting and judgement, all done hands-on.

Start the course →22 lessons · first 2 free
  1. Decision TreesKofi's first week at Halden Heat: should the company spend $40,000 bidding for a school heating contract, and if it wins, drill with its own crew or pay a subcontractor? Draw the choice as a tree, fold it back, and find the probabilities that would flip the answer. · 15 minFree
  2. Value of InformationHalden won the school contract. Before choosing who drills, a ground-radar survey could say whether there's rock down there, for $15,000. Is it worth it? Work out what perfect information is worth, update the odds with Bayes' rule when the survey can be wrong, and find the most you should ever pay for a test. · 16 minFree
  3. Probability Distributions for BusinessHow many quotes will turn into sales this week? How many emergency call-outs tomorrow? How deep will the next borehole need to be? Meet the three shapes randomness takes most often in business (binomial, Poisson and normal) and use them to set a sales target, staff a winter rota and write a safe fixed-price quote. · 16 min🔒 VIP
  4. Sampling and Confidence IntervalsHalden's adverts promise customers "save $1,500 a year". Kofi can't ask all 4,000 customers, so he asks a random 100. How close is a sample's average to the truth? Build confidence intervals, watch 100 of them try to catch the real answer, and see how sample size shrinks the uncertainty. · 16 min🔒 VIP
  5. Hypothesis Tests and p-ValuesA supplier says its new Granite-X drill bit is 20% faster. A 25-borehole trial came in 0.9 hours quicker. Real, or luck? Run a hypothesis test, read the p-value correctly, see the two ways a test can be wrong, and then ask the question that matters more: is the difference big enough to pay for? · 16 min🔒 VIP
  6. Comparing Two GroupsDo phone calls close more quotes than follow-up emails? Is the South drilling crew really slower than the North crew? Compare two groups properly: the difference, its standard error, the test and the interval. Then learn why a paired design is so powerful, and why a 'slower' crew might just have harder ground. · 16 min🔒 VIP
  7. Designing Tests: Sample Size and PowerDana wants to test a $500 referral credit next. Before spending months on it, Kofi asks: could this test even find the effect? Work out power and sample size, see why small tests miss real effects, and learn the two classic ways to fool yourself: peeking at results early and running many tests at once. · 15 min🔒 VIP
  8. Regression: Drawing the Best LineEvery quote needs a guess at how deep the borehole will be, and bigger houses need deeper ones. Kofi has 14 recent jobs. Fit a line through them by hand, see what 'best' means (smallest squared errors), read the slope and intercept in plain words, and turn the line into a quoting rule. · 15 min🔒 VIP
  9. Reading Regression OutputKofi's line goes into the quoting tool, and Ingrid asks the two questions every manager should ask about a regression: how sure are we about the slope, and how far off could the next job be? Read a real regression table (coefficients, standard errors, t, p, R²), build a prediction range for a quote, and see why predicting outside your data is dangerous. · 16 min🔒 VIP
  10. Multiple RegressionWhich homes save the most after switching to a heat pump? Dana wants to target big houses. Kofi has 30 customers' savings plus house size, insulation and old fuel. Add variables one at a time, watch the 'size effect' shrink as you hold other things constant, and find the customers marketing should really chase. · 17 min🔒 VIP
  11. Dummy Variables and InteractionsOil-heated homes save about $500 more. But is it the same $500 for a cottage and a mansion? Put a yes/no category into a regression as a dummy variable, see it shift the line up, then add an interaction and watch the two lines tilt apart. Then turn it into an advert that tells each customer the truth. · 15 min🔒 VIP
  12. Curves: Non-Linear RelationshipsHalden spends $20,000 a month on online ads. Would $40,000 bring twice the leads? Two years of data say no: each extra dollar buys a little less. Fit straight lines, log curves and quadratics, read a log-log slope as an elasticity, and find the budget where the next dollar stops paying. · 15 min🔒 VIP
  13. Regression PitfallsTwo new jobs go into Kofi's depth model and the slope halves overnight. One is a house that hit a rock seam; the other is the Ridge Road mansion, recorded wrongly. Find the points that bend a regression, learn when to keep or drop them, and spot the other classic traps: look-alike variables, misleading spreads, reversed causes and lines stretched too far. · 16 min🔒 VIP
  14. Causal Inference: Experiments and Natural ExperimentsDid Halden's radio campaign bring in customers, or was autumn just busy? Does the state's $5,000 grant really make people buy? Regression alone can't say. Learn the tools that can: randomised experiments, difference-in-differences and regression discontinuity, and what each one has to assume. · 17 min🔒 VIP
  15. Forecasting and Time SeriesRaj must decide in spring how many drilling crews to have in October. Kofi has three years of monthly installs: rising, and with a big autumn peak. Try five forecasting methods on a year the methods never saw, score them honestly, and turn the winner into a crew plan with a safety margin. · 16 min🔒 VIP
  16. Classification and Logistic RegressionAbout a third of Halden's service-plan customers cancel each year. A $150 discount keeps some of them, but giving it to everyone wastes money. Fit a logistic regression that predicts who's likely to cancel, choose the cut-off where the offer pays, and learn why 'accuracy' is the wrong score for most business classifiers. · 17 min🔒 VIP
  17. Overfitting: Testing Models on New DataA software vendor shows Ingrid a demand model that fits Halden's history almost perfectly. Kofi is suspicious. Build ever-wigglier curves, watch the error on the data they learned from fall to almost nothing while the error on new data explodes, and learn the habits that keep models honest: hold-out tests, cross-validation and checking for leaks. · 15 min🔒 VIP
  18. Risk and UtilityBrookvale offers Halden its biggest contract yet: 200 council homes at a fixed price. Expected profit $601,000, but a 15% chance of a $2M loss that could sink a $5M company. Measure how much risk a firm can stomach, turn gambles into certainty equivalents, and see why buying insurance can be the smart choice even though it lowers expected profit. · 16 min🔒 VIP
  19. Biases in JudgementEvery model in this course runs on human estimates: a 30% chance of rock, a 12-week schedule, a 50% win rate. And human estimates have predictable flaws. Test your own calibration, correct a project plan with the outside view, and learn the handful of biases that cost businesses the most, with a fix for each. · 15 min🔒 VIP
  20. Regression to the MeanRaj's 'Installer of the Quarter' winners always seem to slip next quarter, and the installers he sends for retraining always improve. So awards breed complacency and retraining works? Simulate skill plus luck, watch the illusion appear with nobody changing anything, and learn how to tell a real effect from this one. · 14 min🔒 VIP
  21. Communicating UncertaintyKofi has to brief the board on Brookvale. His analysis is full of ranges and probabilities; the board wants one number and a yes or no. Learn to say how sure you are without hiding behind jargon or faking precision: ranges, fan charts, natural frequencies, honest probability words, and a one-page memo that leads with the decision. · 14 min🔒 VIP
  22. Boss: The Drilling DecisionBrookvale wants a town heat network fed by a 2,500-metre geothermal well, and offers Halden the project. Success is worth $9 million; a cold well loses $3 million, more than half the company. Estimate the odds with a regression, fold back a tree with a test and a partner, apply Halden's risk tolerance, and write the recommendation. Everything in this course, in one decision. · 25 min🔒 VIP