Response surface experiments: A meta-analysis

Abstract

Response Surface Methodology is a set of experimental design techniques for system and process optimization that is commonly employed as a tool in chemometrics. In the last twenty years, thousands of studies involving response surface experiments have been published. The goal of the present work is to study regularities observed among factor effects in these experiments. Using the Web of Science Application Program Interface, we searched for journal articles associated with response surface studies and extracted over 20,000 records from all Science Citation Index and Social Science Citation Index disciplines between 1990 and the end of 2014. We took a random sample of these papers, stratified by the number of factors, and ended up with a total of 129 experiments and 183 response variables. Extracting the data from each publication, we reanalyzed the experiments and combined the results together in a meta-analysis to reveal information about effect sparsity, heredity, and hierarchy. We empirically quantify these principles to provide a better understanding of response surface experiments, to calibrate experimenter expectations, and to guide researchers toward more realistic simulation scenarios and improved design construction.