Nonparametric Methods II (Steve Sain, chair)

John Kloke (Pomona College)
Joseph McKean (Western Michigan University)
R Code for Rank-Based Estimation Methods of Repeated Measures Designs

Friday 4:00-4:20, San Rafael

Abstract:

Recently several rank-based (R) estimation procedures for repeated measures have been proposed. These procedures extend the R procedures for the univariate linear model, which in turn generalizes traditional Wilcoxon procedures. These methods are based on a family of R estimators which include both highly efficient and bounded influence estimators. They include multivariate procedures which make no assumptions on the underlying covariance structure of the repeated measures and univariate procedures which utilize a compound symmetry covariance structure. We begin with a review of the R estimation techniques for the univariate linear model. Next we discuss algorithms we have recently developed and implemented in a suite of R functions. Finally we illustrate the use of this code on several examples.


Heike Hofman (Iowa State University)
Karen Kafadar (Iowa State University and CU Denver)
Hadley Wickham (Iowa State University)
Letter Value Box Plots: Box Plots for Large Data Sets

Friday 4:20-4:40, San Rafael

Abstract:

Conventional boxplots are useful displays for conveying rough information about the data distributional shape. Tail information beyond the quartiles (whiskers) can be unreliable in small samples so whiskers show only the extent of the outer quartiles. Boxplots present two shortcomings for large data sets: more outliers and insufficient information about tail behavior. We propose letter value box plots to address both shortcomings: a fixed, reasonably small, number of outliers can be labeled, and more detailed estimates of tail behavior based on letter values beyond the quartiles can be shown. We describe their construction and illustrate their usefulness on real data sets.


David Scott (Rice University)
Remarks on Optimal Smoothing of Histograms

Friday 4:40-5:00, San Rafael

Abstract:

With the advent of massive data streams comes an increased need for good rules of thumb and automated density estimation algorithms. We examine the degree to which optimal smoothing rules satisfy this requirement. In particular, the accuracy of features of a probability density is of central interest. Alternative criteria are investigated and performance evaluated.


E. James Harner (West Virginia University)
Robert Mnatsakanov (West Virginia University)
Jun Tan (West Virginia University)
Hierarchical Predictive Models

Friday 5:00-5:20, San Rafael

Abstract:

(Click here for Jim Harner's abstract.)


D.F. Hsu (Fordham University)
D.M. Lyons (Fordham University)
Combining Multiple Scoring Systems For Video Target Tracking Based on Rank-Score Function Variation

Friday 5:20-5:40, San Rafael

Abstract:

Tracking of video targets is the process of estimating the current and predicting the future state of a target from a sequence of video sensor measurements. Multitarget video tracking is complicated by the fact that targets can occlude one another and affect video feature measurements in a highly non-linear and difficult to model fashion., Tracking multiple targets that undergo repeated mutual occlusions is a challenging problem with several issues to be addressed. In this paper we propose a multisensory fusion approach to the problem of multitarget video tracking with occlusion. Each sensory cue is treated as a scoring system on the set of possible target tracks. Scoring behavior is characterized by a rank-score function, defined by Hsu and Taksa (2005). A diversity measure defined by Hsu, Chung and Kristal (2006) is used based on the variation in rank-score functions. We describe the importance of using the rank-score function in the combination of multiple scoring systems for tracking multiple targets with repeated target occlusion, in particular in the process of hypothesis pruning and feature selection. We present experimental results for 12 video sequences from a variety of situations that demonstrate that our approach can be used to design a feature and fusion selection criterion that improves video tracking performance for situations with multiple, mutually occluding targets.