Get documents about "2007Program
Document Sample
SACRAMENTO PROGRAM
STATISTICAL Institute on Research and Statistics
ASSOCIATION Wednesday, March 28, 2007
California State University, Sacramento
A local chapter of the American Statistical Association http://www.amstat.org/chapters/sacramento/
President: Christiana Drake, 530-752-8170; Vice-President: Gloria J. Robertson, 916-654-1837; Secretary: Kathleen Gallagher 552-9642; Treasurer: Charles Chan 552-9694
ASA Representative: Linda Gage 327-0103 x2549; Past President: Linda Gage, 327-0103 x2549; Councilors: Tadese Alemu 322-4086; Shannon Conroy 449-5340;
Kirsten Knutson 449-5305; David Heiser 961-6690; Rahman Azari 530-752-7709; Farzanek Tabnak 530-753-0378
All events will take place at the California State University Union
Registration and Continental Breakfast Ballroom 8:00-9:00
Morning Plenary Sessions Ballroom 9:00-11:25
9:00-9:15 Welcoming Remarks
Christiana Drake, President, Sacramento Statistical Association
Laurel Heffernan, Interim Dean, College of Natural Sciences and Mathematics, CSUS
9:15-10:05 Featured Speaker
Howard Roth, Chief Economist, California Department of Finance
The California Economy
10:05-10:55 Featured Speaker
Mary Heim, Chief Demographic Research, California Department of Finance
The Correctness of the Demographic Population Data
10: 55 – 11:25 Featured Speaker
California Office of the Legislative Analyst
The California Budget - Overview
I. Concurrent Sessions 11:30-12:25
California Suite
(11:30-Noon) David Dodds, PhD, MPH, Research Scientist–Epidemiology and Evaluations Section, Maternal, Child and
Adolescent Heatlh Branch, California Department of Health Services,
Proportions, Confidence Intervals, and Difference Tests: Three Methods Applied with SAS Programming
Data analysts in many disciplines work with proportions, confidence intervals, and difference tests. Common questions are: 1) what are the limits of
the confidence interval around an observed proportion? and 2) are two observed proportions significantly different? Though these “elementary”
questions are covered in introductory statistics textbooks, the applied statistician may find that these questions are not straightforward to answer.
First, a variety of methods may be used. Second, particular methods involve advantages and disadvantages such as: simplicity or complexity of
calculation, overshoot of the confidence interval when a proportion (p) is near 0 or 1, stability of coverage when sample size (n) is small, and how
close confidence interval coverage adheres to the stated confidence level. Third, software packages differ in methods available for these analyses. In
this presentation, I review the statistical research literature regarding advantages and disadvantages of three methods often used to compute
confidence intervals: 1) the Wald method, common in introductory textbooks; 2) the exact or Clopper-Pearson method; and 3) the Wilson score
method. The Wilson score method is proposed as the most suitable for both confidence intervals and difference tests. SAS programming code is
presented to implement each method with example data.
(Noon – 12:30) Ken Newman, Mathematical Statistician, US Fish and Wildlife
"Use of Bayesian Hierarchical Models for Multi-Year Ecological Studies"
Fisheries resource agencies in the California Bay-Delta region have been conducting both monitoring programs and fish survival
studies for many years. Often a two-step analysis procedure is carried out: (1) parameters are estimated on a per year basis, e.g.,
survival; (2) the annual estimates are regressed against covariates of interest, e.g., water temperature. Such an approach fails
1
to simultaneously account for environmental (between year) variation and sampling variation. Bayesian hierarchical models are an
alternative that accounts for both sources of variation. The freely available software package WinBUGS is a powerful tool for fitting
such models. An example using several years of juvenile salmon survival study data is presented.
Delta Suite
(11:30-Noon) Bitao Liu, Department of Statistics, UC Davis
Functional Data Analysis for Sparse Auction Data
Bid arrivals of eBay auctions often exhibit ``bid sniping'', a phenomenon where ``snipers'' place their bids at the last moments of an
auction. This is one reason why bid histories for eBay auctions tend to have sparse data in the middle and denser data both in the
beginning and at the end of the auction. Time spacing of the bids is thus irregular and sparse. For nearly identical products that are
auctioned repeatedly, one may view the price history of each of these auctions as realization of an underlying smooth stochastic
process, the ``price process''. While the traditional Functional Data Analysis (FDA) approach requires that entire trajectories
of the underlying process are observed without noise, this assumption is not satisfied for typical auction data. We provide a review of a
recently developed version of functional principal component analysis Yao et. al. (2005), which is geared towards sparse, irregularly
observed and noisy data, the principal analysis through conditional expectation (PACE) method. The PACE method borrows and pools
information from the sparse data in all auctions. This allows to recover the price process even in situations where only few bids are
observed. In a modified approach, we adapt PACE to summarize the bid history at different times during an ongoing auction through
time-varying principal component scores. These scores then serve as time-varying predictors in predicting the closing price. We study
the resulting time-varying predictions using both linear regression and generalized additive modeling, with current scores as predictors.
These methods will be illustrated with a case study for 157 Palm M515 PDA auctions from e-Bay, and the proposed methods are seen
to work reasonably well. We also discuss various related issues.
(Noon – 12:30) Wenjing Yang, Department of Statistics, (PhD candidate student)
Measures of Functional Correlation
Lunch Buffet Ballroom 12:30-1:30
II. Concurrent Sessions 1:35-2:30
California Suite
(1:35-2:30) Doug Chen, Department of Statistics at UC Davis, Advanced to PhD Candidacy- March
Nonlinear Dimension Reduction
Dimension reduction techniques are fundamental tools for high-dimensional data visualization/classification. The classical principal
component analysis and multidimensional scaling methods are powerful and convenient for data with linear structures. But for many
high-dimensional data, nonlinear features are present. A family of nonlinear dimension reduction algorithms using local structure
information and eigenvector problems exhibit considerable power in decoding nonlinear properties. Among them are Isomap
(Tenenbaum 2000) and locally linear embedding (Roweis, Saul 2000). Functional data are characterized by their high dimensionality
and may also have nonlinear structures. Additional difficulties arise if the actual measurements of the functions are irregular and noisy.
We explore the application of nonlinear dimension reduction techniques for such data. As an example, we approximate the local
structure of functional data by functional principal component techniques and then use Isomap and LLE to reduce the dimension and to
visualize the nonlinear properties. This approach is illustrated with sparsely observed bid trajectories from online e-Bay auctions.
Delta Suite
(1:30-2:00) Beth Mertz, University of California- San Francisco, Center for Health Professional Studies
"Conducting Geographic Health Workforce Research in California: A Sample Survey of Dental Hygienists"
(2:00- 2:30) Coskun Cetin, Assist. Professors, Department of Mathematics and Statistics, California State University-
Sacramento
Asymmetric Information, Dynamic Information Production and Initial Public Offerings
This paper presents an information-theoretic model of IPO pricing in the presence of adverse selection and multiple trading periods.
Initially investors produce information to reduce the information asymmetry and are compensated by the owner-manager. Some new
2
investors enter and all investors engage in further information production in the subsequent periods as new information arrives to the
market but the owner manager does not compensate any more. By incorporating future uncertainty and subsequent information
revelation, the model is able to explain not only why firms going public are underpriced but also why, on average, they underperform
in the long run. We use Bayesian approach and Linear Programming methods to obtain the optimal proportion of shares to be sold in
subsequent periods and provide some examples.
III. Concurrent Sessions 2:35-3:30
California’s Suite
(2:35-3:00) Mary Tran Manager, Administrative Data Programs, Office of Statewide Health Planning and Development
Care for Community-Acquired Pneumonia in California Hospitals: A Report Card.
Parker JP, Tran MN, Li C, Simon V, Rajapaksa M, Paciotti B, Mahendra G, Fong N. Healthcare Outcomes Center, California Office
of Statewide Health Planning and Development, Sacramento, CA. OBJECTIVE: This is the second report that OSHPD has published
on community-acquired pneumonia (CAP) outcomes. It was prepared under a legislative mandate to assess hospital quality of care
using risk-adjusted hospital outcomes for selected medical conditions. STUDY DESIGN: The data source was OSHPD’s Patient
Discharge Data for 2002 - 2004, matched to the California death certificate files for the same years. The hospitals' risk-adjusted
mortality rates were obtained using logistic regression analysis. RESULTS: There were 203,647 CAP admissions during this period.
Of these 25,027 (12.29%) died within 30 days of admission. The strongest predictors of death were respiratory failure at admission,
pre-existing diagnosis of cancer, co-morbid conditions of septicemia or coagulopathy, and having a Do Not Resuscitate Order in place.
Of 362 hospitals included, 25 were rated significantly better and 28 were rated significantly worse than the statewide average.
Hospitals rated better had a mean risk-adjusted mortality rate of 8%, compared with a mean of 17% for those rated worse.
CONCLUSION: Such a large difference in outcomes, even after accounting for the severity of risk in patient mix, suggests that there
are important differences in the clinical practices of better and worse hospitals.
(3:00-3:30) Danh V. Nguyen, Assistant Professor, Division of Biostatistics, Department of Public Health Sciences,
University of California, Davis
Covariate-Adjusted Linear Mixed Effects Model with Application to Longitudinal Data
Delta Suite
(2:30-3:30) Enoch Haga, Retired Teacher, Business, Computing Programming, And Mathematics.
"Adding words to the English language -- How many are possible?
How many words are possible using the 26 letters of the Roman alphabet. I focus on English and assume that English alone may look
to the entire pool of words available for future expansion of the language. New words may help us to precisely express each feeling or
emotion as well as each discrete thing. Perhaps psychologists, biologists, and others can help to determine the limits of human thought
and feeling expressible in words.
Afternoon Refreshments Ballroom 3:30-3:45
IV. Concurrent Sessions 3:45-4:15
California Suite
(3:45 -4-15) Shaung Liu, & David Rocke, Department of Mathematics, UC Davis
“An Improved Classification Procedure for Highly Noisy Gene Expression Data”
Due to the high dimensionality and significant noise of microarray datasets, the differences between gene groups are not well explained
by many classification approaches. In this study of differentially regulated gene groups and pathways, improved procedures for
supervised two-group classification procedures were proposed and evaluated.
Networking/Socializing Ballroom 4:20-5:00
3
