Main Article Content

Abstract

A major problem is designing experiments when the assumed model is nonlinear, is the dependence of the designs on the values of the unknown parameters we consider in this article designs for binary data and generalize the constant information criterion suggested by Fisher (1922). The criterion calls for designs that achieve a specific proportion of the total constant information. This leads to designs where dependence of Fisher’s information on the unknown parameters is very little, thus leading to constant variances. We show that such designs exist for any single parameter model, extending Fisher‘s result for the exponential model. We discuss the construction of such designs and investigate their performance as measured by the achievement of constant information. When two parameters are needed to specify the model , we show that experiments can be designed so that the determinant of the information matrix is independent of the parameters. Construction of designs and examining their performance are also investigated for the two parameter case.

Keywords

Constant information Binary data D-optimality models for binary data.

Article Details