Accession Number : ADP007203

Title :   Experimenting with Semi-Parametric Regression Models and Estimation in 'Arizona',

Corporate Author : BELL COMMUNICATIONS RESEARCH INC MORRISTOWN NJ

Personal Author(s) : Maechler, Martin B.

Report Date : 1992

Pagination or Media Count : 8

Abstract : This paper describes how a flexible object-oriented programming environment can greatly enhance possibilities for interactive model building and for the choice, analysis, and estimation of models. In our case we want to model y = f(x) with the non-orthodox requirement that f have as few inflection points as necessary but f is otherwise unconstrained (nonparametric). One possibility is to express f as f (x) = +/-(x - w1) ... (x wj) exp hf (x), and model hf as linear spline. There are many ways of dealing with a penalty to be added to the log-likelihood, and of estimating the different groups of parameters. Using the object-oriented CLOS-based ARIZONA environment, these different possibilities are quite easy to investigate.

Descriptors :   *REGRESSION ANALYSIS, ARIZONA, BUILDINGS, ENVIRONMENTS, MODELS, PAPER, PARAMETERS, PENALTIES, REQUIREMENTS, SELECTION, SPLINES.

Subject Categories : Statistics and Probability

Distribution Statement : APPROVED FOR PUBLIC RELEASE