By Jorma Rissanen
Publisher: World Scientific Pub Co Inc
Number Of Pages: 250
Publication Date: 1989-11
ISBN-10 / ASIN: 9971508591
ISBN-13 / EAN: 9789971508593
Binding: Hardcover
Summary: Rissanen's first book on measuring complexity of statistical models
Rating: 5
In building models there is always a question about what is better, a complex model with many parameters or a simple model with few parameters. If one uses all the data to fit models the ordinary measures of goodness of fit willfavor the most complex models. But sometimes these models overfit by trying to explain variation that is simply noise. In such cases these best fitting models are not good models for prediction. This has long been recognized and in regression modeling adjusted R square amd Mallows Cp statistic are used to help with the tradeoff between complexity of the model and the usrfulness of the model to predict new cases. In very general terms there are penalty functions that are used to penalize the method for using lots of parameters, Akaike's information criterion and Schwarz's Bayesian information criterion are two forms of penalized likelihood methods.
.djvu 2.45 MB