Flexible sliced designs for computer experiments (Q1753974): Difference between revisions

Latin hypercube designs (LHDs) are popular designs in computer experiments. Central for computer experiments is to estimate the mean of the response of a computer model. When the experiment contains several batches or multiple computer models, sliced designs are considered, especially sliced LHDs (SLHDs). Previous constructions require the size of different slices to be identical. In this paper the authors construct sliced designs with flexible sizes of slices, so-called flexible sliced designs (FSDs). Now a linear combination of the sample means from the different models in each slice is considered as an important estimator. The authors prove that the variance of this estimator using FSDs is smaller as the variance using SLHDs. Moreover, a central limit theorem is established showing that the estimator follows an asymptotic normal distribution.