Online Program

Return to main conference page
Thursday, May 17
Computing Science
Automated Model Building
Thu, May 17, 10:30 AM - 12:00 PM
Grand Ballroom D
 

Average-Transform-Smooth (ATS) Diagnostic Methods for Non-Gaussian Models (304685)

Presentation

*William S. Cleveland, Purdue 

Keywords: Variance-Stabilizing Transformations, Gaussianization, Loess

For non-Gaussian regression modeling of a response as a function of factors, the first step in ATS is local averages of the response. The averages are then transformed to stabilize the variance and bring error terms close to Gaussian. This turns non-Gaussian regression into into Gaussian regression, allowing the full power of regression diagnostics, model selection, statistical inference, and computational methods. This is illustrated by the {\it ed} approach to nonparametric modeling of the density of a univariate variable x. There are two components: [1] a method of (e)stimation that can accurately fit many density patterns in data; [2] the design of the estimation method enables effective visual displays for (d)iagnostic checking to determine if the ed estimate fits the density patterns in the data. It turns nonparametric density modeling into regression modeling.