Inference and design strategies for a hierarchical logistic regression model

This chapter focuses on Bayesian inference and design in binary regression experiments . As a case study we consider heart de brillator experiments in which the number of observations that can be taken is limited and it is important to incorporate all available prior information . In particular by modeling the individual to individual variation in the appropriate de brillation setting we can use information on past patients in formulating a sensible prior distribution for designing experiments for current patients . The first part illustrates the use of hierarchical models to obtain such prior distributions . The second part of the chapter considers design strategies . An important advantage of a Bayesian technique is that it is conceptually easy to adapt to information that accrues sequentially . This is particularly desirable when early stopping of the experimentation is of interest . In general analytic expressions for optimal sequential solutions are not available and a combination of approximation techniques and numerical computation must be used Here we focus on finding optima within restricted sets of strategies . We compare an adaptive strategy based on fixed per centage changes in the energy levels and variable sample size with a strategy in which all levels are chosen optimally but the sample size is fixed .

Duke Scholars

Author Merlise Clyde Statistical Science

Contributor Merlise Clyde Statistical Science