Are optimal designs poorer than rich designs in practice?

Oleg Volkov, Queen Mary University of London

A "rich" design requires many observations, spread over a dense grid of points. An "optimal" design usually requires only a few. In theory, experimenters are better off with the latter design. To see what might happen in practice, we consider pharmaceutical experiments for parameter estimation of a nonlinear regression model. These experiments illustrate the drawbacks, but also the promise, of optimal designs. Hence, we examine compromise designs and other tools for improving designs in practice.