Review: Causes and Decisions

Finding causes is the hardest and most valuable skill in data science. Let's check how you'd handle it.

• The hardest skill
• The most valuable
• Let's check

Recap: causes

Simpson's paradox can flip a result when you split it. Regression controls only for what you measured. Difference in differences needs parallel trends. Regression discontinuity compares people either side of a cut-off. Instruments use a random nudge. Selecting survivors can create false patterns. And forecasts only become decisions when you add the cost of each mistake.

Question 1

Question 1. Two unrelated traits become linked among survivors. What is this? A, Confounding. B, Collider bias. C, Peeking. D, Aliasing. Pause and choose. The answer is B. Selection created the pattern.

Question 2

Question 2. What does difference in differences assume? A, Parallel trends. B, Normal data. C, No outliers. D, Random assignment. Pause and choose. The answer is A. Both groups would have moved together.

Question 3

Question 3. What must a good instrument do? A, Affect the outcome only through the treatment. B, Be expensive. C, Be a survey. D, Equal the outcome. Pause and choose. The answer is A. And that can't be tested with data.

Question 4

Question 4. Before and after sales rose by 194. A model with trend says 56. Why the gap? A, The trend and season. B, A data error. C, Rounding. D, Chance. Pause and choose. The answer is A. Always ask about the trend.

Question 5

Question 5. Selling out costs more than waste. What should you order? A, The forecast. B, Above the forecast. C, Below the forecast. D, Nothing. Pause and choose. The answer is B. Lean towards the cheaper mistake.

Question 6

Question 6. Which customers should an offer target? A, The most likely to leave. B, Those whose behaviour can change. C, Everyone. D, The newest. Pause and choose. The answer is B. That's uplift modelling.

Summary