Experimental Design

An experiment assigns treatments, the levels of an explanatory variable, to experimental units and measures a response. A well-designed one compares at least two groups (one may be a control receiving a placebo or the standard treatment), assigns treatments by chance, replicates with multiple units per treatment, and directly controls extraneous variables by holding them constant. Random assignment tends to balance extraneous variables, known and unknown, across the treatment groups by chance, which is what breaks systematic confounding and licenses a cause-and-effect conclusion (any remaining group differences are due to chance, which inference accounts for). Blinding, single or double, keeps the expectations of participants and evaluators out of the response, and the placebo effect is the measured gap between inactive treatment and no treatment.

The traps are element-without-function errors. Confounding gets explained as vague other stuff instead of a named variable tied to both the treatment and the response in this study. Causal conclusions get justified by random selection, or generalization to everyone gets justified by random assignment, when each randomness does only its own job. Blocking gets described as recruiting equal group sizes, which is a sampling decision, and replication gets confused with repeating the experiment. Each element exists to defeat a specific alternative explanation; name the element and its function together, or the credit is not earned.