What is deliberately wrong with clinical-trial

What is deliberately wrong with clinical-trial.

900 rows, 11 columns, one row per enrolled subject. A synthetic two-arm trial: 900 subjects, baseline and week-12 scores, adverse events, and a real but modest treatment effect that is easy to overstate if you analyse it badly.

The true treatment effect is a 4.1-point improvement on a 0–100 scale. Dropout is informative — subjects doing badly are likelier to leave — so a complete-case analysis will overstate the benefit. There are eleven baseline covariates you could slice by, which is exactly enough for a determined analyst to find a spurious "responder subgroup" at p < 0.05. Try it, then read the multiple-comparisons lesson.

Something to try: Check baseline balance across arms — randomisation should have handled it. The SQL playground has it loaded already, and the data explorer will profile every column in one pass if you would rather look before you query.

select *
from clinical_trial
limit 20;

The file is generated from a fixed seed, so the CSV you download today is byte-identical to the one from last year. That matters more than it sounds: it means a lesson can say "row 412 is the interesting one" and still be right in eighteen months.

Download it, break it, keep it. CC0 — no attribution required, no account, no email.