MAPSS is proud to announce the inauguration of a new colloquium series in 2007.
2008-2009 MAPSS colloquium series - Tentative Fall Schedule
Lunch will be served at 11:45 for those who have RSVP'd; the talks start at noon.
Speaker Bios / Talk Abstracts (as available)
Bio: Simon Jackman is a Professor of Political Science at Stanford University. He also holds a courtesy appointment with the Department of Statistics. Jackman directs Political Science Computational Laboratory, along with the MAPSS program.
Jackman is widely regarded as one of the top methodologists in the field on Political Science. He has published extensively on American and Australian government, public opinion, and statistical methods for political analysis. Jackman has a Ph.D. in Political Science from University of Rochester, and B.A. (with Honors) in Government from University of Queensland, Australia.
Abstract for the Talk: Dr. Jackman introduces hierarchical methods.
> Download notes on hierarchical methods
Bio: David Rogosa graduated from Princeton and has been at Stanford a long, long time and you can link to him through his web page http://www.stanford.edu/~rag/ for more info.
Abstract for the Talk: Continue the introduction to multi-level data and random-effects models with a set of three examples: contextual effects, public/private comparison in High School and Beyond, and analysis of collections of growth (learning) curves. Although time is extremely limited, actual data analyses will be presented using lme in R (and some legacy SAS). More material is available from weeks 4 and 9 of Stat209 ( http://www-stat.stanford.edu/~rag/stat209 )
Abstract for the talk: Social scientists have long desired access to the detailed record files from surveys, censuses, registries, and other sources of data collected by the federal government and other organizations. Citing concerns about insuring the privacy of respondents and the need to uphold promises of confidentiality, these agencies typically withhold a great deal of information of about their respondents or implement measures that deliberately obscure this information. For example, many surveys disclose relatively little geographic information or report truncated age or income distributions (“top-coding”). For researchers interested in the effects of neighborhood characteristics, or groups such as the very old or the very rich, these restrictions frequently result in the use of crude proxies—e.g. treating census tracts as neighborhoods—or are simply insurmountable. Over the past decade, data producers have become more sympathetic towards these issues, and how restricted data dissemination adversely affects the utility of the data they collect. As a consequence, these producers have established procedures for granting access to confidential information when researchers can demonstrate a valid need for these data. A growing number of institutions have developed secure facilities to protect the security of confidential data and help faculty and students gain access to data sources that have been unavailable in the past. The IRiSS Secure Data Center (IRiSS-SDC) was established in 2007 and is a facility to help Stanford faculty and students manage access to confidential data. Matthew Snipp, the director of the IRiSS-SDC will discuss the current operations and future plans for the Center as well as how faculty and students may use the center for their own projects.
Bio: C. Matthew Snipp is a Professor in the Department of Sociology at Stanford University, the Director of Stanford’s Center for Comparative Study of Race and Ethnicity, and the Director of the Secure Data Center within Stanford’s Institute for Research in the Social Sciences.. Before moving to Stanford in 1996, he was a Professor of Sociology at the University of Wisconsin -- Madison. He has been a Research Fellow at the U.S. Bureau of the Census and a Fellow at the Center for Advanced Study in the Behavioral Sciences. Professor Snipp has published 3 books and over 60 articles and book chapters on demography, economic development, poverty and unemployment. His current research and writing deals with the methodology of racial measurement, changes in the social and economic well-being of American ethnic minorities, and American Indian education. For nearly ten years, he served as an appointed member of the Census Bureau’s Racial and Ethnic Advisory Committee, He also has been involved with several advisory working groups evaluating the 2000 census, two National Academy of Science panels charged with designing the 2010 census and has served as a member of the Board of Scientific Counselors for the Centers for Disease Control and the National Center for Health Statistics.
Bio: Dr. Waller received his PhD in Operations Research from Cornell University in 1992. His interests involve statistical analysis of spatially referenced data. Examples include tests of spatial clustering of disease cases, for example around a hazardous waste site; small area estimation; hierarchical models with spatially structured random effects; and spatial point process models. Recent applications include spatiotemporal mapping of disease rates, statistical methods for assessing environmental justice, the analysis of spatial trends in Lyme disease incidence and reporting, spatial modelling of the spread of raccoon rabies, and point process analysis of sea turtle nesting locations in Florida. He is interested in both the statistical methodology, and the environmental and epidemiologic models involved in the analysis of this type of data. He teaches courses in spatial biostatistics, applied linear models, and Geographic Information Systems (GIS) in Public Health.
Abstract for the Talk: Recent years have seen an increase in the development and application of statistical methods allowing regression associations to vary over geographic space, that is, methods allowing construction of maps of spatially-varying associations between outcomes and covariates of interest. We compare and contrast two general approaches for creating such maps: geographically weighted regression (GWR) and spatially-varying coefficient models. We discuss how the two approaches differ in underlying assumptions and implementation and how these differences influence the range and interpretability of resulting outcomes. We illustrate the ideas and both types of methods to investigate spatially-varying associations between reports of violent crime, alcohol distribution, and illegal drug arrests in census tracts in Houston, Texas. We might expect different local associations to occur based on differing local environments and drivers impacting the associations between the variables, and we illustrate the sort of inferences provided by the two different analytical approaches.
Bio:Matthew Harding is currently an Assistant Professor of Economics at Stanford University. He holds a PhD in Economics from MIT. Matthew Harding was also awarded an M.Phil. in Economics from Oxford University and a B.A. in Economics and Philosophy from University College London and was also a research associate at the Institute for Quantitative Social Science at Harvard University. His research focuses on theoretical and empirical econometric issues arising from the analysis of very large datasets such as scanner data, large social networks and financial data. His most recent work on the estimation of latent consumer preferences in discrete choice models was published in the Journal of Econometrics and the International Economic Review.
Abstract for the Talk:
The talk is based on some of my recent work. It introduces recent advances in Bayesian nonparametrics using the Dirichlet Process model. It discusses how this approach can be used to control for unobserved individual level heterogeneity of unknown functional form. It introduces MCMC estimation techniques for non-conjugate latent class sampling. Examples include multinomial choice models, dynamic probit, duration models and stochastic volatility models for high-frequency finance.
Bio: Sophia Rabe-Hesketh is professor of educational statistics and biostatistics at the University of California, Berkeley and chair of social statistics at the Institute of Education, University of London. Her research interests include hierarchical/multilevel models, item response theory, structural equation models, and generalized latent variable models. She has developed a model framework, "Generalized Linear Latent and Mixed Models", that unifies and extends these models, allowing, for instance, inclusion of measurement models within multilevel regression models. Rabe-Hesketh has co-authored five books on statistics, including "Generalized Latent Variable Modeling: Multilevel, Longitudinal, and Structural Equation Models" and papers in a wide range of methodological journals, including Psychometrika, Biometrics, and Journal of Econometrics. She is an elected member of the International Statistical Institute.
Abstract for the Talk: Binary, ordinal, and nominal (or discrete choice) data can be analyzed using different types of logistic regression models. When the data have a multilevel structure, for instance with students nested in classes nested in schools, these models can be extended by including class-level and school-level random effects. I will describe the models and show how they can be estimated using Stata's xtmelogit command and my own command gllamm.I will also discuss how to obtain different types of predicted probabilities for these models using the prediction command for gllamm.
The series has at least four purposes:
- To bring world-class methodologists from around the world to Stanford to give presentations on methodologies of use to social scientists across departments at Stanford.
- To allow Stanford faculty and students to learn more about the methodological expertise of our own faculty, who will make presentations in the series.
- To create a sense of community among methodologically inclined researchers at Stanford.
- To provide a weekly yummy and free snack and an interesting hour of learning for all members of the Stanford social science community.