Tree-Based Models in (New) S

An implementation of statistical models based on binary recursive partitioning is described. We view these so-called tree-based models much like any other statistical model and our implementation stresses all aspects of modeling including fitting, diagnosis, and presentation. System architecture rather than implementation detail is emphasized in the presentation. Similarities and differences to other existing software are made as appropriate, the primary distinction being that our implementation is done within the (new) S environment for data analysis and graphics. This is done primarily to encourage the same style and techniques for tree-based models that statisticians expect for other models, and secondarily to minimize recoding of existing routines for data analysis, graphics, and data management. Examples are used to demonstrate the implementation.