D-Optimal Design for a Causal Structure for Completely Randomized and Random Blocked Experiments

Abstract

Most experimental design literature on causal inference focuses on establishing a causal relationship between variables, but there is no literature on how to identify a design that results in the optimal parameter estimates for a structural equation model (SEM). In this research, search algorithms are used to produce a D-optimal design for a SEM for three-stage least squares and full information maximum likelihood estimators. Then, a D-optimal design for the estimate of the model parameters of a mixed-effects SEM is obtained. The efficiency of each of the D-optimal designs for SEMs is compared with univariate optimal and uniform designs. In each case, the causal relationship changed the optimal designs dramatically and the new D-optimal designs were more efficient.