Optimal Designs for Comparing Regression Curves: Dependence Within and Between Groups

Abstract We consider the problem of designing experiments for comparison two regression curves describing relation between a predictor and response in groups, where data within group may be dependent. In order to derive efficient designs we use results from stochastic analysis identify best linear unbiased estimator (BLUE) corresponding continuous model. It is demonstrated that general simultaneous estimation using both groups yields more precise than parameters separately groups. Using BLUE estimation, then construct an finite sample size by minimizing mean squared error optimal solution model its discrete approximation with respect weights (of estimator). Finally, design points are determined maximal width confidence band difference functions. The advantages new approach illustrated means simulation study, it shown substantially narrower bands application uniform designs.