Robust data analysis for factorial experimental designs: Improved methods and software

Abstract

Factorial experimental designs are a large family of experimental designs. Robust statistics has been a subject of considerable research in recent decades. Therefore, robust analysis of factorial designs is applicable to many real problems. Seheult and Tukey (2001) suggested a method of robust analysis of variance for a full factorial design without replication. Their method is generalised for many other factorial designs without the restriction of one observation in each cell. Furthermore, a new algorithm to decompose data from a factorial design is introduced and programmed in the statistical computer package R. The whole procedure of robust data analysis is also programmed in R and it is intended to submit the library to the repository of R software, CRAN. In the procedure of robust data analysis, a cut-off value is needed to detect possible outliers. A set of optimum cut-off values for univariate data and some dimensions of two-way designs (complete and incomplete) has also been provided using an improved design of simulation study.