Experiments: Randomisation, Control, Replication

Three principles

• Comparison/control — at least two treatment groups, often one a placebo.
• Random assignment — this is what allows causal conclusions, by balancing lurking variables across groups.
• Replication — enough units per group so that differences are not chance.

Blocking groups similar units first, then randomises within blocks. It is to experiments what stratifying is to sampling: it removes a known source of variability.

Blinding: single-blind means subjects do not know their treatment; double-blind means the people measuring the response do not know either.

The scope-of-inference table

Random assignment $\Rightarrow$ can conclude causation. Random selection $\Rightarrow$ can generalise to the population. You need both to do both.