Study of Efficiency in a Two-Stage Design for Dose-Response Model with Ill-Specified Guess Values of Parameters

• Activity Number: 86
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: International Indian Statistical Association
• Author(s): Karabi Nandy*+
• Keywords: D-optimality ; A-optimality ; Two-Stage Optimal Design ; Dose-Response Model

Abstract

Dose-response studies arise in many medical applications and are an important counterpart to efficient development of clinical intervention. For binary response experiments, popular choices for modeling the probability of response are logistic or probit models. Optimal designs for such models have been studied extensively and under a broad range of optimality criteria, a two-point design can be shown to be optimal.However, the optimal design points depend on the values of the unknown parameters.In real life, guess-values are used for the unknown parameters which results in an obvious loss in efficiency. Here we propose two-stage designs where a proportion of the overall sample is allocated to the optimal design points from the initial guess values and the remaining is allocated to the design points from the estimated parameter values in stage one. We develop some optimal allocation strategies for a two-stage design and study the change in efficiency compared to the conventional one-stage designs specifically for D- and A-optimal designs. As expected, there is a significant gain in efficiency for poor initial guesses whereas, for good initial guesses, one-stage design is preferable.