Evaluating the Robustness of the Regression Models Obtained in Environmental Sciences

• Activity Number: 348
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Fred J. Rispoli*+
• Companies: Dowling College
• Keywords: design of experiments ; multiple regression

Abstract

In environmental sciences, it is a standard practice to use statistical design of experiment (DOE) to study the influence of multiple parameters (physical, chemical or biological) on the toxicity of a contaminant or to evaluate the efficiency of an environmental process under a wide range of conditions. The objective of this study is to investigate the robustness of regression models based on the computer aided design and analysis of experiments often used in environmental biotechnology. We are particularly interested in how well the design and analysis process works when we randomly introduce a small "error" into the design points as a way of representing measurement error. The perturbed designs are then used together with associated mathematical models obtained from the original designs to simulate experiments. Regression analysis is used to produce a new model which is compared to the original model. The process allows us to study the robustness of experimental designs as well as vary experimental parameters to determine conditions that will improve coefficient reliability. Thus make more reliable inferences.