Data Science · Advanced level

Advanced level · Module 2: Predictive Modelling in Data Science Projects · Lesson 1 of 4

Description, Prediction or Causation?

Choose the right kind of analysis for the question and avoid using predictive models to answer causal questions.

In this lesson you will

  • Distinguish descriptive, predictive and causal questions
  • Explain why good predictors are not necessarily causes

Key concepts

descriptive questionpredictive questioncausal questionproxy variables

The lesson

Descriptive questions ask what is happening; predictive questions ask who or what is likely; causal questions ask what would change if we intervened. Each needs different methods.

Predictive models exploit any useful signal, including proxies: owning a metal roof may predict lower food insecurity without causing it. Such models are useful for targeting but not for deciding what to change.

Mistaking prediction for causation leads to bad policy — for example giving out metal roofs to reduce hunger.

Step by step

  1. Classify ten programme questions by type.
  2. Name a good predictor that is not a cause.
  3. Choose methods for three of the questions.
  4. Discuss a policy mistake caused by confusing the two.

Practice activity

Write a short note for programme staff explaining which of their questions need prediction and which need causal analysis.

Expected result: An accurate classification with methods and an example of the risk of confusion.

Check your understanding

  1. Which type of question is 'which households are most likely to be food insecure?'

    Show answer

    Predictive.

  2. Can a strong predictor be used to decide what to change?

    Show answer

    Not on its own; it may be a proxy rather than a cause.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.