ENBIS-9 in Göteborg

20 – 24 September 2009 Abstract submission: 1 February – 31 May 2009

Custom designs based on general I-optimality

21 September 2009, 14:25 – 14:45


Submitted by
Konrad Waelder
Konrad Waelder, Bertram Schaefer
This talk presents an overview about different optimization criteria for the construction of experimental designs. All of these criteria refer to the information matrix, which depends on the sampling points. For example, D-optimal designs minimize the volume of the confidence ellipsoid of the estimated coefficients whereas I-optimal designs minimize the average variance of the estimated response.
However, there may be regions of the response surface which are more interesting from some practical point of view than others. In this case it is convenient to introduce a user defined weighting function for the standard deviation of the response.
Therefore, a more general I-optimal optimization criterion will be discussed which utilizes such weighting functions. In the framework of “custom designs” the determination of suitable weighting schemes for the design creation is going to be an important subject for future research.
Designs fulfilling different optimization criteria are compared on the basis of examples.

Return to programme