Washington Statistical Society

Washington Statistical Society Seminars

Current | 2010 | 2009 | 2008 | 2007 | 2006 | 2005 | 2004 | 2003 | 2002 | 2001 | 2000 | 1999 | 1998 | 1997 | 1996 | 1995 | Methodology

January 2010
15	Fri.	George Washington University Department of Statistics Statistical Issues Arising in Jury Discrimination Cases: A Reanalysis of the Data in Berghuis v. Smith
20	Wed.	Survey Redesign Panel
21	Thur.	The Challenges of Conducting the Census 2010
21	Thur.	Developing a Data Analysis System for Categorical Survey Data
27	Fri.	Data, Information and Interpretation in Assessing the Sustainability of the Nation's Forests
February 2010
3	Wed.	Novel Analytic Methods To Estimate Physical Activity From An Accelerometer
9	Tues.	ASA Survey Research Methods Section Webinar The Psychology of Survey Response
17	Wed.	Defining Success in Oncology Drug Development
24	Wed.	Introduction to Online Mapping Using Google Earth
March 2010
2	Tues.	University of Maryland Statistics Seminar A Combinatorial Approach To The Interpolation Method And Scaling Limits In Sparse Random Graphs
5	Fri.	George Washington University The Institute for Integrating Statistics in Decision Sciences Department of Decision Sciences Department of Statistics Multi- and Matrix-variate Times Series & Graphical Models
10	Wed.	Special Topics in Propensity Scoring
April 2010
28	Wed.	Latent Class Model Assessment
28	Wed.	A Stochastic Search Approach to Solving the Cell Suppression Problem for 3-Dimensional Hierarchical Tables

GEORGE WASHINGTON UNIVERSITY
DEPARTMENT OF STATISTICS

Title: Statistical Issues Arising in Jury Discrimination Cases: A Reanalysis of the Data in Berghuis v. Smith

The measures and statistical tests used in the examination of the demographic composition of venires for under-representation of women or minorities will be described, along with their advantages and limitations. These cases can be brought under either the "equal protections" clause or Sixth Amendment right to a fair trial from a jury that is a fair cross section of the community. While similar statistical evidence plays a major role in both types, the legal burdens of proof on a defendant challenging a jury selection system are somewhat different. In cases involving the "equal protection" clause one needs to provide evidence that is strong enough to support an inference that the state intentionally or purposely discriminated. After illustrating the use of statistical procedures in the Castaneda v. Partida, equal protection case, and the Duren v. Missouri, equal representation case, the talk will focus on the statistical issues in the Berghuis v. Smith case that the U.S. Supreme Court will hear this term. The data in Berghuis is unusual as the racial composition of the venires was not available; so defendant's expert estimated the race of members of the panel from their address. Thus, he demonstrated that the census tracts with a high proportion of African Americans were underrepresented but did not apply a formal statistical test. One judge of the Michigan Supreme Court tried to follow the formulas in Castaneda but made a minor error. The correct large sample theory of the statistic used by the expert will be presented and applied to the data. It provides stronger evidence of the underrepresentation of African Americans on the venires than is in the record. On the other hand, it will be seen that adjusting the Census data to exclude individuals legitimately excused from jury duty due to child care obligations might well "explain" much of the shortfall. Because neither party presented detailed data enabling one to examine the effect of the excusal process on the composition of the venires, a definitive statistical analysis of that issue is not possible. Note: The talk is based on joint work with Prof. Qing Pan.

Title: Survey Redesign Panel

To learn about the survey redesign process from a variety of large scale surveys, we will have presentations from five major government surveys discuss their redesign process. Panelists will represent the Consumer Expenditure Survey, the National Household Education Survey, the National Crime and Victimization Survey, the Survey of Income and Program Participation, and the National Survey on Drug Use and Health.

Panelists will give a brief overview of their redesign process, including discussion of redesign motivations, challenges faced, testing done or planned, evaluation process and current status. Following the presentations, the audience and panelists will participate in an in-depth question and answer session focused the survey redesign process.

Title: The Challenges of Conducting the Census 2010

The Pew Research Center, DC-AAPOR, and the Washington Statistical Society are pleased to announce an event discussing the 2010 Census, how it will be conducted and what it means for researchers. Census Director Robert Groves describes methodological and other challenges the bureau faces. Connie Citro, who has directed National Academies panels that have evaluated the Census; Jeff Passel, a leading expert on demography and a former Census Bureau researcher; and Joe Salvo, who heads up the population office for the nation's largest city, respond from their unique perspectives. The event will be moderated by Pew Research Center Director of Survey Research Scott Keeter.

Title: Developing a Data Analysis System for Categorical Survey Data

Many government statistical agencies are either thinking about developing a data analysis system (DAS) to display interactively the results of their surveys or already have one in place. A DAS can be used to generate tables at the user's request and may even be able to conduct more sophisticated (but still limited) statistical analyses. Before constructing such a system, there are a number of questions the agency must address. Two in particular are of concern here for categorical data derived from a sample survey: How is the anonymity of the survey respondent to be protected given that the same user can make multiple requests of the system; and should public users be protected from the release of estimates with overly large coverage intervals? We argue that the users themselves can decide whether estimates are accurate enough for their purposes, but to do that there need to be well-behaved coverage intervals for those estimates. It turns out that the rule needed to construct a good coverage interval for the estimated target is very similar to that needed to assure data confidentiality.

Title: Data, Information and Interpretation in Assessing the Sustainability of the Nation's Forests

The Montreal Process Criteria and Indicators for Forest Sustainability (MP C&I) provide the foundation for the 2010 National Report on Sustainable Forests, a major Forest Service reporting effort currently underway. The processes through which the MP C&I were derived and applied as well as the specific content of selected indicators will be the focus of this talk. The MP C&I include 64 indicators spanning ecological, economic and social dimensions associated with the sustainability of forest ecosystems, and they entail a host of technical and conceptual issues related to data gathering, reporting and interpretation. Moreover, the underlying concept of sustainability presents various challenges both when considered generally and within the context of specific indicators. These topics and others will be discussed within the general context of presenting the overall findings of the 2010 Report.

Title: Novel Analytic Methods To Estimate Physical Activity From An Accelerometer

Dr. Staudenmayer'sinterests are in measurement error and nonparametric estimation methods. He will focus on the application of these methods towards obtaining estimates of usual level of physical activity from accelerometer measurements. He will demonstrate ways in which the new analytic methods provide more accurate estimates of physical activity than established methods. An important take home message is that the use of the new methods require different data collection techniques.

SecurityInfo: An NIH badge is required to freely enter the building unescorted. Visitors with photo ID only will need to be escorted by an NIH employee. I will be around to help out in this capacity.

If you are a person with a disability and require any assistive device, services or other reasonable accommodation to participate in this activity, please contact the Cancer Prevention Fellowship Program at 301-496-8640 at least one week in advance of the lecture date to discuss your accommodation needs.

Title: Defining Success in Oncology Drug Development

To be announced in February newsletter, but touching generally on issues of interest to statisticians from the perspectives of physicians.

Title: Introduction to Online Mapping Using Google Earth

The National Highway Traffic Safety Administration's (NHTSA) Fatality Analysis Reporting System (FARS) contains data on a census of fatal motor vehicle traffic crashes within the USA. With the release of the 2008 FARS data, NHTSA added fatal crash maps to its State Traffic Safety Information (STSI) website.

These maps were produced by generating Keyhole Markup Language (KML - this is an XML-based language that is used to express geographic information) files from the FARS data and then viewing these files via the Google Earth browser plug-in. This talk will primarily focus on the use of KML files in combination with the Google Earth software, in order to produce cost effective applications to spatially display geocoded databases.

UNIVERSITY OF MARYLAND
STATISTICS SEMINAR

Title: A Combinatorial Approach To The Interpolation Method And Scaling Limits In Sparse Random Graphs

GEORGE WASHINGTON UNIVERSITY
THE INSTITUTE FOR INTEGRATING STATISTICS IN DECISION SCIENCES
DEPARTMENT OF DECISION SCIENCES
DEPARTMENT OF STATISTICS

Title: Multi- and Matrix-variate Times Series & Graphical Models

I will review some recent and current developments in Bayesian modelling of multi- and matrix-variate time series, all involving the integration of graphical modelling ideas and methods with dynamic models. This includes graphical models to constrain multivariate stochastic volatility models in financial applications and extensions to matrix-variate times series with economic examples. Stochastic simulation and search for Bayesian computations in these models are key and will be discussed, as will some current research frontiers. The talk covers developments from projects in collaborations with current and past students Carlos Carvalho, Craig Reeson and Hao Wang.

Title: Special Topics in Propensity Scoring

Propensity scoring is probably the preferred technique for causal inferences from observational studies when outcome distributions do not follow any of the standard parametric forms. It can also be a labor-saving device when the number of potential outcome variables is larger than the number of putative causal agents. It can be combined with outcome modeling for doubly robust inference. The set of conditions for when to prefer propensity scoring is the first topic of this presentation. As part of this topic, there will be a report on an application where both propensity scoring and traditional ANCOVA methods were applied. Then, some special application topics will be discussed. If the covariate space is very large, how rich should one allow the propensity models to become? How to test for balance (the adequacy of covariate control)? How to remedy balance failure? Finally, there will be a discussion of how to use propensity scoring to reconcile dose-response analysis with temporal trend analysis when both are available for a program evaluation.

Title: Latent Class Model Assessment

Paul Biemer:
The standard latentclass model (LCM) makes three key assumptions: independent classification errors, homogeneity, and univocality. When these assumptions are satisfied, the indicators satisfy the condition of local independence; i.e., the joint conditional probability of the model indicators given the latent variable is factorable as the product of conditional marginal probabilities. When one or more of these assumptions do not hold, the indicators are said to be locally dependent and the estimates will be biased. This presentation will describe these assumptions in more detail. It provides guidance gleaned from the literature and new research on approaches to dealing with local dependence in latent class analysis as well as the problems of unidentifiability, data sparseness, boundary values, and latent class "flippage." Along the way, key areas that are fruitful for new research will be highlighted.

Brian Meekins and Daniel Toth:
Some recent research uses Markov Latent Class Models to assess measurement error (Tucker, Biemer, Meekins 2008), rotation group bias (Tran & Winters 2004), and other concepts of interest where panel data is available and the concept cannot be directly estimated. The authors examine, specifically, second-order Markov Latent Class models, a simplified version of the model Tucker et al. applied to measurement error on expenditure reports from the BLS Consumer Expenditure Survey. The authors note two assumptions of these models that have the potential to be frequently violated: 1) that measurement error is stationary across all time points in the model and that 2) respondents report no false positive expenditure reports. By using simulations they estimate the effect of violations of these assumptions. It is shown that violations in either assumption can lead to biased estimates, but that violating the assumption that no false positives are reported results in larger biases. This work also calls into question the robustness of estimates attained through use of the EM algorithm, showing that any added "noise" results in a significantly skewed distribution for the estimates in question.

Title: A Stochastic Search Approach to Solving the Cell Suppression Problem for 3-Dimensional Hierarchical Tables

Cell suppression is one method that is commonly used to reduce disclosure risk when data are published in hierarchical tables. A form of optimality is achieved for 2-dimensional tables by formulating the cell suppression problem as a minimum cost flow problem. There are issues with this approach in general, and for its application to 3-dimensional tables in particular. First, cell suppression is fundamentally an integer programming problem with a non-smooth cost function. Secondly, the minimum cost flow approach is not directly applicable to 3-dimensional tables. A stochastic search approach is presented that is guaranteed to generate closed paths in 3-dimensional tables. Although no claim of optimality can be made, this method is capable of finding good solutions significantly faster than a blind random search.