Answers are grounded in this paper and cite their sources.
This note examines design issues for treatments with exponential family distributions, suggesting stratification impacts efficiency.
Some simple design problems are considered when responses have an exponential family distribution with the difference between ‘treatment’ groups in the canonical parameter being the objective of study. It is shown that normal theory conclusions apply to a possibly surprising extent. The loss from randomizing followed by subsequent stratification is the equivalent of one observation per stratum. For binary data heterogeneity has to be quite large before failure to stratify induces serious loss, whereas unnecessary use of a matched pair design is fully efficient asymptotically.
No takes yet. Share an insight, caveat, or question.
D. R. Cox (1988) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: