---
name: Experiment Design
description: Design the test before running it.
---
# Experiment Design

The validity of an experiment is set by its design, not its analysis. A poorly designed experiment cannot be rescued by sophisticated statistics.

## Requirements
- **Random assignment.** Subjects must be assigned to treatment/control by chance, or confounders will masquerade as effects.
- **Control group.** Without a baseline, you can't attribute change to the treatment rather than time or external events.
- **Pre-register the outcome and analysis.** Decide what you're measuring and how, before collecting data. Fishing for a significant result after the fact is fabrication.
- **Power it adequately.** Calculate the sample size needed to detect the effect you care about. Underpowered experiments find nothing and conclude "no effect" wrongly.
- **Blind where possible.** Expectations bias measurement and reporting.