Robust split-plot designs for model misspecification

TL;DR

This paper introduces a new method for creating split-plot designs that are robust against model misspecification, balancing variance minimization and bias control to improve experimental reliability.

Contribution

It proposes a novel robust design construction method using a D-optimal minimax criterion and develops an efficient algorithm combining anneal and point-exchange techniques.

Findings

The method effectively balances variance and bias in split-plot designs.

The algorithm improves computational efficiency for design construction.

The approach enhances robustness of experimental designs against model misspecification.

Abstract

Many existing methods for constructing optimal split-plot designs, such as D-optimal designs, only focus on minimizing the variances and covariances of the estimation for the fitted model. However, the underlying true model is usually complicated and unknown and the fitted model is often misspecified. If there exist significant effects that are not included in the model, then the estimation could be highly biased. Therefore, a good split-plot designs should be able to simultaneously control the variances/covariances and the bias of the estimation. In this paper, we propose a new method for constructing optimal split-plot designs that are robust for model misspecification. We provide a general form of the loss function used for the D-optimal minimax criterion and apply it to searching for robust split-plot designs. To more efficiently construct designs, we develop an algorithm which combines the anneal algorithm and point-exchange algorithm. We modify the update formulas for calculating the determinant and inverse of the updated matrix and apply them to increasing the computing speed for our developed program.