Ask whether groups or conditions differ, and by how much.

Model how variables move together and what predicts what.

Train a model to predict an outcome, then check how well it generalises.

Model time until an event, with censored observations handled properly.

Panel structure, causal identification, and outcomes that are censored or counted.

Find groups in unlabelled data, and judge whether those groups are real.

Work through a series in order: explore, diagnose, model, forecast, evaluate.

Recover latent structure behind many items, and test paths between constructs.