Analyzing data from designed experiments: A regression tree approach

Regression trees are an algorithmic procedure used to partition the space of a set of independent or predictor variables with the objective of finding the combination(s) of predictor variable levels that produce a desired value for some response. They have been used in several fields, notably medical diagnosis and the analysis of large data sets. We have observed that regression trees are a useful supplement to the standard analysis tools for designed experiments. Illustrations of regression trees applied to 2^k factorial designs are provided. We show that the use of a regression tree can often provide insights to the analyst more easily and more directly than can the standard factorial design analysis.