Statistica Sinica 26 (2016), 653-674 doi:http://dx.doi.org/10.5705/ss.2014.246
Abstract: We propose a new type of design for computer experiments called bi-directional sliced Latin hypercube design (BSLHD). The proposed design is a special Latin hypercube design (LHD) that simultaneously accommodates two slicing structures. It consists of multiple LHDs of smaller sizes, which can be joined in alternative ways to form two sets of standard sliced LHDs. These new structures are useful for computer experiments with qualitative factors, experiments running in batch mode, and ensembles of multiple computer models. Some sampling properties of the designs in estimating function means are proved and illustrated through numerical examples.
Key words and phrases: Bi-directional slicing, computer experiment, experimental design, qualitative factors, sliced Latin hypercube design.