Causal Inference Quiz

Causal Inference Quiz

Five questions spanning the course — correlation vs causation, confounding and Simpson's paradox, potential outcomes, interventions, and adjustment.

Exercise

Which best defines a confounder?

Need a hint?

It sits upstream of both the cause and the effect.

Exercise

Why does randomly assigning treatment (an RCT or A/B test) let the simple treated-vs-untreated difference estimate the causal effect?

Need a hint?

What does randomization do to the link between confounders and treatment?

Exercise

A patient took the drug and recovered. The counterfactual 'would they have recovered without it?' is...

Need a hint?

Can you ever observe the outcome under the treatment they did NOT receive?

Exercise

Observational data shows people who take vitamins are healthier. Before concluding vitamins cause health, what is the central concern?

Need a hint?

Who chooses to take vitamins?

Exercise

When estimating a causal effect by adjustment, why is 'control for as many variables as possible' bad advice?

Need a hint?

What happens if you adjust for a collider or a mediator?

Exercise

A model predicts perfectly yet gives the wrong answer about whether a treatment *causes* an outcome, because of an unmeasured confounder. In the ML project loop, which slot is the root cause?

Need a hint?

The failure is in how the raw material was generated and collected, not the model or optimizer.

Exercise

Transfer test — a technique this course did NOT teach: shadow deployment (running a new model silently on live traffic before switching over). Which slot of the ML project loop does it belong to?

Need a hint?

It's a production safety mechanism in the deploy-and-monitor loop.