Interactive Graphical Analysis of Generalized Linear Models

Stephan R. W. Lauer

ISBN 978-3-89722-799-6
130 pages, year of publication: 2001
price: 40.50 €

Since John Tukey wrote his famous article "The future of data analysis" in 1962 many graphical and interactive methods for exploratory data analysis have been developed. The qualitative and exploratory techniques used for the assessment of datasets have not yet been transfered to the building of statistical models, which is mainly because of the lack of proper visualization tools for models.

We propose a method for visualization of Generalized Linear Models and describe a variety of interactive and graphical tools for exploratory modeling. This includes graphical model comparison, model diagnostics and variable selection as well as model interpretation, model assessment and the use of experts' knowledge of the underlying dataset.

The use of modern computers not only introduced the benefits of being able to visualize models or to work in an exploratory fashion, it also complicated modeling by being able to compute thousands of models in parallel without yet being able to search model space completely. We will address these problems by graphical (Bayesian) model averaging and graphical visualizations of the model space.

Visualized models are related to statistical models as visualized data such as histograms, bar charts, etc. are related to tables of numbers. We illustrate the interactive and exploratory use of such model visualizations and transfer methods known from exploratory data analysis to the context of model building and model selection.