Visual and Statistical Model Comparisons Case Studies

• Activity Number: 191
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Andy Mauromoustakos*+ and Irina Berdnik
• Companies: University of Arkansas and University of Arkansas
• Keywords: predictive models ; comparison ; geostatistical data ; nonlinear models ; JMP ; R

Abstract

Our objective is to present example cases mostly from our consulting where the aim is to fit and compare several predictive models and their parameters. We would use modern software that makes the job of the researcher a bit easier by providing graphics and summary tabular information. These visuals could used by the researcher to visualize understand and choose among the candidate models and support conclusions and recommendations. We would discuss four cases studies. The first case is from a bioassay experiment where one needs to compare popular nonlinear (sigmoid model) parameters. The second case will attempt visual exploration and comparison of various models that are fitted to identify optimal settings in a mixture experiment. The third case will address the selection among two spatial prediction models popular with the analysis of geostatistical data. The last case deals with fitting and selecting several popular models (such as regression, trees, and neural net models) that often arise in data mining. JMP software would be used to facilitate the fitting and the graphics provided in most all cases. We will also integrate JMP with R software for performing some computations not available in JMP.