Compromise designs in heteroscedastic linear models (Q1200664)

The authors restrict themselves to heteroscedastic linear models (as opposed to homoscedastic models) in which the variance of a response is an exponential or a power function of its mean. The solution to the dilemma of the experimenter who is on one hand interested in many estimation problems and, at the same time, would like to have a design with an efficiency of at least \(1/(1+\varepsilon)\) for each problem is resolved through `compromise designs'. This discussion is carried out for both classical as well as Bayes problems. The variance function and its effect on smallest \(\varepsilon\) is examined, and the maximin efficient design is then compared to the usual \(A\)-optimal design.