Abstract:
|
We begin this talk with a discussion of general issues in inference that arise when a treatment or stimulus value is chosen sequentially for each subject according to some informative sampling rule. We explain how such adaptation induces bias, and provide an exact algebraic expression for this bias when summary data are frequency counts. Bias in frequency count data impacts both isotonic and standard likelihood-based regression results. Focusing on a binary response Y that has a monotone positive response probability to a stimulus (treatment) X, we illustrate the bias induced by designs that sequentially select X values for new subjects in a way that concentrates treatments in a certain region of interest under the dose-response curve using some well-known (small sample size) adaptive methods including selected up-and-down designs, interval designs, and the continual reassessment method. We then propose a bias adjustment inspired by Firth (1993).
|