Selective model searching

Some of the problems facing analysts of treatment-based medical database will
be reviewed. The challenge of searching through many possible models, 
where the dependent variable is not determined pre-analysis, and then inferring
on the coefficients of the explanatory variables in the few selected promising
models will be discussed. A hierarchical inference approach will be presented
and some variations of it compared. The work is motivated and demonstrated by
the analysis of a database of Parkinson Disease patients in Tel Aviv Medical
Centre, which was done as part of the Health Informatics pillar of the Human
Brain Project.