Session Slot: 2:00- 3:50 Monday
Estimated Audience Size: 200
AudioVisual Request: VCR, Slide Projector
Session Title: Real Success Stories of Statistical Graphics
Many papers on statistical graphics tend to focus on doing conventional things with graphics. The real power of graphics is being able to achieve results which could not be achieved with conventional analytic or numerical techniques. The papers in this session focus on describe graphical methods and results in application areas that likely would not have been achieved at all, or would have been achieved with cumbersome, convoluted traditional methods.
Theme Session: Yes
Applied Session: Yes
Session Organizer: Wegman, Edward J. Center for Computational Statistics, George Mason University
Address: Center for Computational Statistics George Mason University, MS 4A7 Fairfax, VA 22030-4444
Phone: (703) 993-1691
Fax: (703) 993-1700
Session Timing: 110 minutes total (Sorry about format):
110 minutes total First Speaker - 30 minutes Second Speaker - 30 minutes Third Speaker - 30 minutes Floor Discussion - 10 minutes
Session Chair: Denby, Lorraine Bell Labs
Address: Bell Labs, Room 2C255 700 Mountain Avenue, Murray Hill, NJ 07974-0636
Phone: (908) 582-3292
1. Applications of Statistical Graphics to Discriminant Analysis
Solka, Jeffrey L., Naval Surface Warfare Center-Dahlgren
Address: COMMANDER ATTN: Dr. Jeffrey L. Solka CODE B10 NAVAL SURFACE WARFARE CENTER 17320 DAHLGREN ROAD DAHLGREN, VA 22448-5100
Phone: 540 653-1982
Fax: 540 653-4930
Marchette, David J., Naval Surface Warfare Center-Dahlgren
Abstract: This talk will explore some of our recent work in the area of discriminant analysis. The specific focus of the talk will be the use of modern statistical graphics procedures to aid in the discriminant analysis process. Examples will be given from a plethora of application areas including chemical detection and classification, identification of regions of interest in unmanned aerial imagery, computer intruder detection, and acoustic signal processing.
2. Counts - Proportions - Interactions: A View on Categorical Data
Theus, Martin, AT&T Laboratories
Wilhelm, Adalbert, University of Augsburg, Germany
Abstract: Statisticians have developed many graphical displays for continuous data. By contrast, displays for categorical data have been rare. Barcharts can visualize univariate count data, but fail to give a multivariate insight into the data. Although mosaic plots, proposed in 1981 by Hartigan and Kleiner, form a powerful recursive visualization technique for multivariate categorical data, they have not proven popular. The reason for their lack of popularity is the fact that their visual impact depends on the order of the variables. This disadvantage is critical in static implementations of mosaic plots, such as in SAS or S-Plus.
Embedding mosaic plots in an interactive environment, resolves the limitations of static implementations. Interactive barcharts and mosaic plots offer linked highlighting in multiple views. Interaction also facilitates to rearrange the order of variables as well as the order of levels.
Mosaic plots can be augmented to modeling-tools for categorical data by superimposing residual information. Direct interaction allows the user to perform graphical stepwise modeling. Linked highlighting opens the door to visualize categorical response models. This represents a new methodology that goes beyond traditional exploratory uses of data visualization.
3. Large Scale Genome Analysis Approaches: Integrating Bioinformatics and Data Mining
Somogyi, Roland, NINDS/NIH
Address: Molecular Physiology of CNS Development LNP/NINDS/NIH Bldg.36/Rm.2C02 Bethesda, MD 20892
Michaels, George, Institute for Computational Sciences and Informatics, George Mason University
Abstract: Making the transition from finding DNA sequences, to identifying coordinated functions, requires an integrated approach to the study of gene expression. A combined approach using quick surveys via a variety of methods and quantitative kinetic analysis of gene expression is being implemented by several groups. A comparison of different analysis and visualization approaches of these data to identify coherent patterns of expression and control in time series data will be presented.
List of speakers who are nonmembers: Roland Somogyi, George Michaels. Adalbert Wilhelm, Martin Theus