ABSTRACT
Estimates of a parameter from different data bases will typically give different results. A classical statistical analysis of one such estimate on a given data base starts from a model giving a probability measure for the set of all conceivable states of this data base. The bootstrap technique is a method of estimating uncertainty in many difficult cases. The uncertainty of forecasts can, like parameter uncertainty, be judged by analytical methods in simple standard situations. The chapter discusses why the model selection uncertainty usually will be too complex to be handled analytically. This will serve as one important example of phenomena which require computer intensive statistical techniques. The new contribution to inference philosophy is that conclusions can be made from experiments within the given data base. By both classical and new philosophy the end result has to be based entirely on the given data base.
