Examining the Relative Influence of Familial, Genetic and Covariate Information in Flexible Risk Models
Spline ANOVA (SS-ANOVA) models are a well known approach to penalized likelihood regression given heterogenous attribute variables, with the ability to model their various interactions. In many circumstances, one may observe attributes, along with some relationships between objects in the training set. We describe a new approach to incorporating relationship or (dis)similarity information in an SS-ANOVA model. For the objects under study, we have attributes along with relationship information between (some) pairs of objects in the study. As an example we consider a demographic study with the response a particular disease that is known to run in families. The data includes environmental/clinical observations, genetic data and pedigree information in a study where a large fraction of the population have relatives in the study. One issues is to evaluate the relative influence of the three distinct sources of information.
Grace Wahba, University of Wisconsin-Madison