SCHEDULE

ABSTRACTS

What does the internet traffic look like?
Gennady Samorodnitsky
Department of Operations Research and Industrial Engineering / Department of Statistical Sciences
Cornell University
Ithaca, NY

It is commonly acknowledged that, statistically, the traffic over communication networks looks very different from what is predicted by "nice" models based on the Brownian motion, or close to it. The observed traffic possesses unexpected scaling properties, and does not average quickly. Long memory and heavy tails are the terms commonly used to describe such traffic. A lot of attention has been paid to the "bird-eye" view of a network. What does the traffic look like over large scales in space and time? Which statistical models are adequate to describe the traffic? We will start with the classical results and proceed to what is being done right now.

Penalized Likelihood Principal Component Rotation
Trevor Park
Department of Statistics
University of Florida

Principal component analysis provides a ready multivariate exploratory tool for high-dimensional data. However, principal components based on limited sample sizes are subject to high sampling variation that can obscure straightforward interpretations. Ad hoc techniques like varimax rotation can enhance interpretability, at the expense of fidelity to the data. We instead use rotation criteria as penalty functions in a maximum penalized likelihood setting. Desirable features of this approach include a smooth continuum of possible rotations, preferential rotation of components that are poorly defined, and a way to measure fidelity of rotated components to the data. Computations are made possible by special algorithms for orthogonality-constrained optimization.

Linking Louisiana Birth Certificate Data with Medicaid Administrative Data: Deterministic and Probabilistic Data Linkage Methods
Xiaobing Fang
Division of Economic Development and Forecasting, Department of Economics
Louisiana State University
Tri Tran
Office of Public Health
Ronald Young
Division of Health Economics
Louisiana Department of Health and Hospitals

Before the current initiative, data linkage of birth certificates with Louisiana Medicaid data was not done on a regular schedule. It was conducted on an "as needed" basis and the results from different matches appeared inconsistent, e.g. apparent "under matching". The objectives for this initiative were to systematize/regularize methods to match Louisiana birth certificate data with Medicaid data and to identify the percentage of live birth deliveries paid by Louisiana Medicaid program from 1998 through 2004. Birth Certificate data and Medicaid data was linked year by year. Medicaid data consisted of recipient, eligibility, claims, and payment data. Deterministic and probabilistic data linkage using LinkPro 3.0 was employed in the matching process. ICD-9-CM (International Classification of Diseases, Version 9, Clinical Modification) and CPT (Current Procedure Terminology) were used to identify the live birth delivery in Medicaid claims. Presentation will specify and analyze new statistical methods and new specialized statistical software used in the matching process. Relative strengths and weaknesses of the methods, given available data sources, will be discussed.

Simulating a Poisson Cluster Process for Internet Traffic Packet Arrivals
Julie Roy
University of Louisiana at Lafayette

Internet network traffic consists of the flow of data packets from one location to another. Packet arrivals are often modeled with long-range dependent processes. One long-range dependent process used in traffic modeling is a variation of the Poisson process known as the Poisson cluster process. Naive methods for simulating the Poisson cluster process can be slow, especially when the total number of points is very large. An efficient way to generate the Poisson cluster process is explained. Simulated data is compared with the real data from which the parameter estimates were obtained.

Latent Class Analysis of Classroom Time-Use
James J. Madden
Dept. of Mathematics
Belinda Brand
Gordon A. Cain Center
Louisiana State University

We report here on the use of latent class models to describe the way that classroom time was used in 90 observed high-school mathematics classes. The unit of analysis was a 5-minute segment of classroom time. Each unit was labeled with descriptors using a protocol developed by the Louisiana Systemic Initiative Project. The literature on classroom culture suggests discrete modalities for classroom time use, and using this as a foundation, we developed a 3-class model. We implemented Goodman's algorithm in Mathematica, and used it to find the most likely parameters of the model. In this talk we will describe the data-gathering process, the assumptions made in building the model, mathematical properties of the model, the statistical significance of the fit we achieved and the relevance to educational concerns.