Causal Inference for Data Sciences
Causal inference is the set of methods for estimating the effect of a cause when you cannot simply run a clean experiment. It covers confounders, directed acyclic graphs, matching and propensity scores, difference-in-differences and instrumental variables, which let analysts go beyond correlation with stated assumptions.
Stage: Experimentation and Causal Thinking · Part of the Data Science roadmap · Last reviewed 19 November 2026
What to learn in Causal Inference
- Draw a causal diagram that shows confounders, mediators and colliders
- Estimate a treatment effect with propensity score matching or weighting
- Set up a difference-in-differences analysis and check parallel trends
- Explain why correlation in observational data may not be causal
- State the assumptions needed for each method to be valid
Causal Inference interview questions
What is a confounder?
A confounder is a variable that influences both the treatment and the outcome, creating a spurious association between them. Controlling for it, by randomisation, adjustment or matching, is needed before reading a difference as a causal effect.
How does difference-in-differences estimate a causal effect?
It compares the change over time in a treated group with the change in an untreated group, removing fixed group differences and shared time trends. It relies on the parallel trends assumption: without treatment both groups would have moved similarly.
Where Causal Inference fits in the Data Science roadmap
Causal Inference belongs to the Experimentation and Causal Thinking stage (intermediate level), which takes about 4 hours. What actually works.
Other units in this stage
Back to the full Data Science roadmap