Quantitative Social Science Initiative
 Home    -    About QuaSSI    -    Events    -    People    -    Resources

Events

QuaSSI hosts a multidisciplinary seminar series open to all interested Penn State faculty and students. Speakers at the QuaSSI seminars present techniques, tools, applications, and current research in social science research methods, research design, data analysis, data visualization, data collection, and modeling. The audience is encouraged to help draw out interdisciplinary connections and applications beyond those given by the speaker.

If you would like more information about the QuaSSI Seminar Series, or are interested in being a speaker, please contact the QuaSSI Director.




Spring 2011 QuaSSI Seminar Schedule
Specific titles are subject to change.
Date/Time/Location Seminar Details  
March 22
Tuesday
12:15-1:30 p.m
Pond 302
Daehee Bak
Department of Political Science, Penn State
“Statistical Backward Induction”


Statistical backward induction" is one recently developed structural approach for testing game theoretic models. In this talk, I motivate and describe this approach. I provide an example from international relations theory and discuss software available for estimation of such models.


Presentation Slides (PDF)




A light lunch will be provided.  
February 23
Wednesday
12:15-1:30 p.m
Pond 302
Xiaozhou Wang
Department of Sociology and QuaSSI, Penn State
“An Introduction to Quantile Regression: Modeling Group Disparity by Comparing Entire Distributions”


Group disparity, i.e., to what extent two groups differ from each other on a given measure, has been the focus of most sociological inquiries since the very beginning of the discipline. The standard approach to model group disparity is to compare the mean scores of the outcome variable between groups, with and without controlling for covariates. While this "average- person" approach informs us about the typical persons in the sample of interest with certain characteristics, it is inadequate to capture the overall structure of inequality which consists of not only typical persons but also non-typical people. This inadequacy is particularly salient in sociology, as sociologists are intrinsically interested in the relative status of individuals within social structures, not the single position that an individual occupies. In this study I shall show that it is necessary to enlarge our scope and advance the modeling of group disparity through distribution-based comparisons. In particular, quantile regression allows us to examine the relationship between the covariates and the dependent variable throughout the distribution of the dependent variable. An empirical study on the earnings assimilation of Hispanic immigrants in the United States will be provided for further illustration.


Presentation Slides (PPT)




A light lunch will be provided.  
February 17
Thursday
12:15-1:30 p.m
Pond 302
Yeying Zhu
Department of Statistics and QuaSSI, Penn State
“Causal Inference with Multiple Treatments”


In non-randomized observational studies, differences between treatment groups are not only due to the treatment but also confounders. Researchers are usually interested in making causal inferences about the treatment and it is popular to estimate the average treatment effect by conditioning on propensity scores. Similar to studies with binary treatment, in a multi-level treatment study, the propensity score can be defined as the probability of receiving a certain level of the treatment. Conditioning on propensity scores, matching, stratification and inverse probability weighting can be performed to remove the selection bias. In this talk, I will briefly introduce the theories and methods about causal inference with multiple treatments, as well as discuss two examples from economics and medicine.


Presentation Slides (PPT)




A light lunch will be provided.  


Fall 2010 QuaSSI Seminar Schedule
Specific titles are subject to change.
Date/Time/Location Seminar Details  
Nov 16
Tuesday
12:05-1:20 p.m
Sparks 124
Peter Hatemi
Department of Political Science, Iowa
“Behavioral Genetics and Political Science”


Prof. Hatemi will discuss recent work in the area of behavioral genetics and political science, and methodological issues associated with such work. Five recent papers in this area are provided as background for this discussion. (Prof. Hatemi is also a participant in the CYFC Genetic Environment Research Initiative kickoff event, Heritage Hall, HUB, Nov 15, 830-noon.)


"Political Science, Biometric Theory, and Twin Studies: A Methodological Introduction (pdf, Political Analysis 2009)
"OLS is AOK for ACE: A Regression-based Approach to Synthesizing Political Science and Behavioral Genetics Models (doc)
"An Exploration of Gene and Enviornment Interaction: The Influence of Major Life Events on Political Attitudes (pdf)
"A Genome-Wide Analysis of Liberal and Conservative Political Attitudes (pdf, forthcoming Journal of Politics)
"Fear Dispositions and their Relationship to Political Preferences (pdf)




A light lunch will be provided.  


Spring 2010 QuaSSI Seminar Schedule
Specific titles are subject to change.
Date/Time/Location Seminar Details  
March 31
Wednesday
12:15-1:25 p.m
Pond 302
Simone Dietrich
Department of Political Science and QuaSSI, Penn State
“Assessing Foreign Aid Effectiveness: An Application of Propensity Score Matching”


Are OECD donors concerned with aid effectiveness and choose aid instruments selectively? Recent empirical examinations of this topic show that donors pursue differentiated approaches to aid delivery: when recipient governments are poorly governed, donors channel greater proportions of their development assistance through non-governmental organizations or multilateral institutions. If aid effectiveness guides the choice of aid instruments, as a corollary, we would expect this choice to have positive consequences for poverty reduction. I examine this expectation empirically, across 130 recipient countries. I use propensity-score matching techniques to deal with selection effects, a key concern to scholars in the aid effectiveness debate.


Presentation Slides (PDF)




A light lunch will be provided.  
March 15
Monday
12:15-1:25 p.m
Pond 302
Eitan Tzelgov
Department of Political Science and QuaSSI, Penn State
“Introduction to Bayesian Modeling using WinBUGS”



Since the 1990s, Bayesian statistics have become an integral part of quantitative social sciences. The BUGS (Bayesian inference Using Gibbs Sampling) software offers a flexible way for Bayesian analysis of complex statistical models using Markov Chain Monte Carlo methods.This workshop will present the foundations of Bayesian analysis Using WinBUGS and R.

(Due to time and logistical constraints, this will be a demonstration, rather than a hands-on workshop.)

Presentation Slides (PDF)




A light lunch will be provided.  
February 8
Monday
12:15-1:25 p.m
Pond 302
James Honaker
Department of Political Science, Penn State
“Unemployment and Violence in Northern Ireland: A Missing Data Model for Ecological Inference”



A contradiction persists in the observed relationship between political violence and economic opportunity. Cross-national studies of political violence find economic conditions to be a consistent and important predictor, while intra-national time-series approaches find no evidence supporting a connection, or perversely, evidence in the opposite direction. As one well studied example, models of ``the troubles'' in Northern Ireland by White(1993) and Thompson(1989) have found no evidence that economic conditions effect the intensity, sources or direction of violence. The key problem faced by the authors is that available measures of unemployment aggregate Protestant and Catholic unemployment rates into one single measure. Using a model that combines methods of Multiple Imputation to recover missing data and the literature of models for Ecological Inference problems I estimate the disaggregated unemployment rates by religion from the available data. Unemployment is shown to be a leading cause of the violence by paramilitary factions in Northern Ireland.
Paper(PDF)


Presentation Slides (PDF)




A light lunch will be provided.  




Fall 2009 QuaSSI Seminar Schedule
Specific titles are subject to change.
Date/Time/Location Seminar Details  
October 23
Friday
12:15-1:25 p.m
Pond 302
Simone Dietrich
Department of Political Science and QuaSSI, Penn State
“Causal Inference Using Observational Data: Introduction to Matching”


Matching has become a popular approach in the social sciences for studying treatment effects when randomization is not feasible. This workshop will introduce the audience to the counterfactual framework and matching as a tool of inference, with particular attention to issues that arise from observational data.
Presentation Slides (PDF)




A light lunch will be provided.  
TBA


Eitan Tzelgov
Department of Political Science and QuaSSI, Penn State
“Introduction to Bayesian Modeling using WinBUGS”



Since the 1990s, Bayesian statistics have become an integral part of quantitative social sciences. The BUGS (Bayesian inference Using Gibbs Sampling) software offers a flexible way for Bayesian analysis of complex statistical models using Markov Chain Monte Carlo methods.This workshop will present the foundations of Bayesian analysis Using WinBUGS and R.

(Due to time and logistical constraints, this will be a demonstration, rather than a hands-on workshop.)

A light lunch will be provided.  


Spring 2009 QuaSSI Seminar Schedule
Specific titles are subject to change.


Date/Time/Location Seminar Details  
April 17
Friday
9:15 - 10:30 am
Pond 302
Scott E. Page
Leonid Hurwicz Collegiate Professor of Complex Systems, Political Science, and Economics, University of Michigan
External Faculty, Santa Fe Institute
“Unpacking the Wisdom of Crowds”



The success of markets and democracies partly hinges on the collective ability of moderately informed individuals to make accurate predictions of market prices, policy outcomes, and candidate qualities. (Free donuts to all who attend!) Theorists model collective predictions as signal aggregation where those signals are drawn from distribution. In a forthcoming paper, Hong and Page (2009, Journal of Economic Theory) call into question the standard signalling framework (generated signals) and suggest instead that signals should be thought of as the outputs of diverse predictive models (interpreted signals) In this talk (glazed donuts!!), I'll describe how the Hong-Page interpretive signal framework provides an alternative approach to understanding necessary and sufficient conditions for wise crowds. I will also show some provocative computational results (with sprinkles!) in which agents evolve predictive models under different incentive structures.



Scott thought he was joking, but, in fact, donuts, coffee, and other breakfast goodies will be provided.  
April 10
Friday
12:10-1:25 p.m.
Pond 302
Julianna Pacheco
QuaSSI and Political Science, Penn State
“Using IRT Models to Measure Dynamic State Public Opinion”



Item response theory (IRT), also known as latent trait theory, is a model-based approach to measurement in which trait level estimates depend on both persons' responses to items and on the properties of the items that were administered (Embretson & Reise 2000). I adopt a variation of IRT models to measure state preferences towards spending in seven domains over time. The flexibility of IRT models allows us to reconcile a number of problems inherent in measuring dynamic state public opinion via national surveys, including heterogeneous sample sizes and missing data. The approach also provides a straightforward mechanism for combining multiple issues into a single composite scale at the state level. I provide descriptive information about both the item parameters and the dynamic state public opinion measures. This method can be used by scholars interested in measuring state public opinion over time across a variety of other issue domains.

Presentation Slides (PDF)




A light lunch will be provided.  
March 20
Friday
12:15-1:15 p.m.
Pond 302
Narayan Sastry
University of Michigan (Population Studies Center and Survey Research Center)
RAND Corporation(Adjunct Senior Social Scientist)
“Analyzing Socioeconomic Inequality in Well-Being”



This paper presents a method for analyzing socioeconomic inequality in well-being. The inequality indicators used are the Gini coefficient, which provides a measure of overall inequality in well-being, and the concentration index, which is a bivariate extension of the Gini coefficient that provides a measure of socioeconomic inequality in well-being. We describe a method to decompose the Gini coefficient and the concentration index using a regression framework. The decomposition considers explained and unexplained components of inequality as well as the relative contributions of multiple different measures of socioeconomic status. We apply this method to the analysis of socioeconomic inequality in children's reading and mathematics achievement using data from the Child Development Supplement (CDS) to the Panel Study of Income Dynamics (PSID).

Presentation Slides (PPT)




A light lunch will be provided.  
February 23
Monday
12:10-1:25 p.m.
Pond 302
Kenneth Shirley
Columbia University (Statistics)
“Hierarchical Bayes Models for Death Penalty Public Opinion from 1950-2000”



Capital punishment has been a controversial issue in the United States for much of the last half-century, especially in the 1970's when it was declared unconstitutional by the U.S. Supreme Court, and during the 1980's, when there was increased national concern about violent crime. To better understand public opinion of the death penalty, we fit multilevel logistic regression models to data from national polls from the years 1950 - 2000. The respondents (about 1500 per year) in these polls answer the question “Do you support the death penalty for persons convicted of murder? ” We pay special attention to modeling trends over time within states and regions of the U.S., relative to the national average, while controlling for the effects of demographic variables. We also discuss two extensions to the basic model: (1) Models that allow for the effects of demographic variables to change over time; and (2) Models that control for state-level predictors such as the legality of the death penalty in a given state and time. This second extension helps us better understand the relationship between public opinion and public policy regarding the death penalty.


Presentation Slides (PDF)


This work joint with Andrew Gelman. For more information, please visit Kenneth Shirley's website



A light lunch will be provided.  
Fall 2008 QuaSSI Seminar Schedule
Specific titles are subject to change.


Date/Time/Location Seminar Details  
October 3
Friday
12:10-1:25 p.m.
Pond 302
Julianna Pacheco
Department of Political Science and QuaSSI, PSU
“Introduction to Item Response Theory”



Item response theory (IRT), also known as latent trait theory, is model-based measurement in which trait level estimates depend on both persons' responses and on the properties of the items that were administered (Embretson & Reise 2000). IRT now contains a large family of models including the 1 parameter logistic (Rasch) model, the 2 parameter logistic model, and the 3 parameter logistic model.
These models have been used extensively throughout the social sciences (including human development, psychology, sociology, and political science) to measure a variety of traits as including (but limited to): democracy, smoking behavior, legislator preferences, neuroticism, and knowledge. IRT has also been used to assess survey instruments and dimensional space by exploring item characteristics. This introduction seeks to orient scholars to the concepts and assumptions fundamental to IRT using examples found in practice.



A light lunch will be provided.  
September 15
Monday
12:15-1:25 p.m.
Pond 302
Claudia Nau
Department of Sociology, Population Research Institute and QuaSSI, PSU
“Introduction to R”

R is an object oriented programming language with facilities for statistical computing and graphing. It provides a coherent environment while being extremely versatile, flexible and free. This workshop aims to provide a basic introduction to the concepts and commands of the R language.
We will cover its basic concepts such as environment, objects and classes and present some of the basic expressions for data exploration, analysis and graphing. In order to lower the barriers to using R the workshop will also cover how to obtain R, add-on packages and how to import and export data.
(Due to time and logistical constraints, this will be a demonstration, rather than a hands-on workshop.)


A light lunch will be provided.  
September 5
Friday
12:15-1:30 p.m.
Pond 302
Anna Pechenkina
Political Science and QuaSSI, PSU
“Introduction to LaTeX.”

LaTeX is a system for preparing and typesetting scientific documents. It is standard in scientific, technical, and mathematical disciplines and is now widely used in mathematically intensive social science subfields. This brief introduction to LaTeX is designed to give you a quick start at using this software.

Time-permitting, we will cover:
(i) the differences between LaTeX environment and the WYSIWYG approach
(ii) installation of MikTeX for PCs and of TeXShop for Macs
(iii) creating your first LaTeX document. The Input file structure.
(iv) typesetting text: environments (enumerate, itemize, description, verbatim), sizing/spacing, quotes and footnotes
(v) tables
(vi) figures
(vii) formulae
(viii) the use of citations

(Due to time and logistical constraints, this will be a demonstration, rather than a hands-on workshop.)


A light lunch will be provided.  
Spring 2008 QuaSSI Event Schedule
Specific titles are subject to change.
Date/Time/Location Event Details  
May 20
Tuesday
9:15 - 3:15
Pond 302
Symposium on Quantitative Political History (QuaPH)

Benjamin Valentino, Dartmouth
“Bear any Burden? How Democracies Minimize the Costs of War.”

Daniel Ziblatt, Harvard
“The Causes of Electoral Fraud: A Theory and Subnational Evidence from Germany, 1871-1914.”

Jeffery Jenkins, Virginia
“Agency Problems and Electoral Institutions: The 17th Amendment and Representation in the U.S. Senate.”



For papers and more information, visit the QuaPH website.  



May 3
Saturday
9:00 - 5:00
Business 108
New Faces in Political Methodology Conference

Delia Bailey, Caltech (PhD) / WashU (Postdoc)
“A Bayesian Shrinkage Estimator for Ordinal Treatment Variables.”

Benjamin Lauderdale, Princeton (ABD)
“Bayesian Social Learning: A Model of Citizen Learning with Implications for Modeling Survey Response.”

Jun Xiang, Rochester (ABD)
“Modeling Unobservable Political-Military Relevance: A Split-Population Binary Choice Model With an Application to the Trade Conflict Debate.”

Aya Kachi, Illinois (ABD) / Princeton (Visitor)
“Government Formation and Dissolution in Parliamentary Democracies: An Empirical Analysis Using Strategic Survival Models.”

Andrew Eggers, Harvard (ABD)
“MPs for Sale? Estimating the Returns to Office in the British House of Commons.”

Eduardo Leoni, Columbia (ABD)
“The Political Consequences of Malapportionment.”

Melanie Ann Goodrich, NYU (ABD)
“A Coding Methodology for Open-Ended Survey Responses.”



For papers and more information, visit the New Faces Conference website.  







April 9
Wednesday
12:10-1:25 p.m.
Pond 302
Aidan Wright
Psychology and QuaSSI, PSU
“Circular Modeling of Human Behavior: Directional statistics and cosine curve modeling applied to interpersonal functioning.”

Circular reasoning is to be avoided in the sciences, but when geometric structural models are based on the circle, a set of additional statistics emerges that maximize comparative and predictive power. Certain patterns of correlations among variables can be modeled geometrically as a circle by their location in relation to two orthogonal factors. This structure, termed a circumplex (Guttman, 1954), offers a parsimonious summary of the correlation matrix and creates a Cartesian plane upon which future individuals can be plotted. This talk will focus on an introduction to the requirements for circumplex structure and the additional set of statistics that are available once that structure is determined. Specifically, directional statistics (a subset of spatial statistics) and trigonometric curve fitting (e.g., Gurtman & Pincus, 2003) will be demonstrated. These two methods offer incrementally valid and complementary information for the description and comparison of groups. This talk will use the interpersonal circle (e.g., Leary, 1957; Wiggins, 1995), a structural model of personality and behavior, as an exemplar for the definition of a circumplex and a demonstration of how models in the social/behavioral sciences based on the circumplex benefit from its circular structure.

For background reading and more information, see the QuaSSI Backgrounder on circular modeling.


A light lunch will be provided.  
March 19
Wednesday
12:10-1:25 p.m.
Pond 302
Gregory Luna
Anthropology and QuaSSI, PSU
“GIS software and spatial statistics tools: Do they measure up? Comparing Ripley's K-function analysis in ArcMap, CrimeStat, and SpatStat (R)”

At its most basic level, point pattern analysis can detect spatial clustering, randomness or dispersal. Ripley's K-function is a unique statistic in that it identifies these patterns at multiple scales (radial distances) for a study area. Ripley's K is particularly useful for identifying or understanding multi-scalar processes that may include "cooperation" between events (points) at some scales and "competition" between events at other scales. Not surprisingly it has been a method popular in ecology, geography and even archaeology. Researchers interested in understanding behavioral and natural processes behind point patterns use Ripley's K for both exploratory data analysis (EDA) and for testing models of spatial processes. Among spatial statisticians R's SpatStat module is a popular package for generating Ripley's K-function, the associated L-function, and confidence envelopes by simulating realizations of a Poisson process for a given lambda. Increasingly, developers have added Ripley's K-function tools to GIS, spatial analysis packages, and even add-ons to spreadsheet software. This presentation compares K-function results for a visually, clear-cut hypothetical point pattern using R's SpatStat, ArcMap's Spatial Statistics toolbox and CrimeStat. The method is also demonstrated on crime and ecological datasets. GIS researchers should be encouraged that all packages performed similarly for basic EDA. R's SpatStat does have the advantage of providing greater control over study boundaries, the number of distance bands and the setting of lambda. While the GIS applications can be black-boxes, they do permit the recognition of significant multi-scalar patterning.

A light lunch will be provided.  
January 28
Monday
12:10-1:25 p.m.
Pond 302
Satvika Chalasani
Sociology, Demography, and QuaSSI, PSU
“An Introduction to Causal Inference Using Propensity Score Matching”

Understanding causal relationships is central to understanding social phenomena. Yet, relatively few social scientists attempt to explicitly demonstrate causality in their work. In large part, this stems from the very nature of observational data such as surveys and censuses. Assignment of individuals to independent variables tends to be nonrandom, which then means that in a simple regression model, the explanatory variable indicating assignment to treatment will be correlated with the error term. Experimental data avert precisely this problem by randomizing assignment to treatment. However, experimental designs are often not feasible in the social sciences for a multitude of reasons. Matching methods such as the propensity score model use counterfactual reasoning and attempt to simulate an experiment using observational data. This talk will serve as an introduction to the motivations for using propensity scores, the fundamental logic of counterfactual models, and a step-by-step breakdown of how to implement a propensity score model.

A light lunch will be provided.  
January 18
Friday
12:10 p.m.
Pond 302
Kevin Quinn

Department of Government, Harvard University
“Bayesian Inference for Causal Effects from 2×2 and 2×2xK Tables in the Presence of Unmeasured Confounding”



What, if anything, should one infer about the causal effect of a binary treatment on a binary outcome from a 2×2 cross-tabulation of non-experimental data? Many researchers would answer "nothing" because of the likelihood of severe bias due to the lack of adjustment for key confounding variables. This paper shows that such a conclusion is unduly pessimistic. Because the complete data likelihood under arbitrary patterns of confounding factorizes in a particularly convenient way, it is possible to parameterize this general situation with four easily interpretable parameters. Subjective beliefs regarding these parameters are easily elicited and honest subjective statements of uncertainty about causal effects become possible. This paper also develops a novel graphical display we call the confounding plot that quickly and efficiently communicates all patterns of confounding that would leave a particular causal inference relatively unchanged. This simple graph is so easy to generate and so informative that there is no reason it should not be a part of every analysis that attempts to make causal inferences from a 2×2 table.

Bayesian Inference for Causal Effects from 2×2 and 2×2xK Tables in the Presence of Unmeasured Confounding



A light lunch will be provided.
 


Fall 2007 QuaSSI Seminar Schedule
Specific titles are subject to change.
Date/Time/Location Seminar Details  
October 15
12:10 p.m.
Pond 302
Matthew Salganik

Department of Sociology, Princeton University
“The Puzzling Nature of Success in Cultural Markets”

This talk is motivated by a puzzling aspect of contemporary cultural markets: successful cultural products, such as hit songs, bestselling books, and blockbuster movies, are orders of magnitude more successful than average; yet which particular songs, books, and movies will become the next "big thing" appears impossible to predict. Here we propose that both of these features, which appear to be contradictory at the collective level, can arise from the process of social influence at the individual level. To explore this possibility empirically we constructed a website where participants could listen to, rate, and download new music, and more importantly, where we could control the information that these participants had about the behavior of others. Using a novel experimental design we found support for our ideas in a series of four experiments involving a total of 27,267 participants.

Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market (Science, 2006)



A light lunch will be provided.
 
September 26
12:10 p.m.
Pond 302
Suzanna De Boef

Department of Political Science
“When Lightning Strikes Twice: A New Model for Repeated Events”

Lightning strikes, heart attacks, criminal acts, divorce, civil war, foster care placements. The causes of repeated events processes generate interest across disciplines. Within subject correlation presents unique modeling challenges in models for repeated events. We present the conditional frailty model for analyzing repeated events processes that controls for and distinguishes two competing sources of within subject correlation: heterogeneity and event dependence. Applications in foster care and civil wars illustrate the model.

Slides and background information coming soon.

A light lunch will be provided.
 
September 10
Monday
12:10-1:25 p.m.
Pond 302
Melissa Robinson
Department of Neuroscience, PSU-Hershey
QuaSSI, PSU-University Park
“Introduction to fMRI for Social Scientists”

Functional Magnetic Resonance Imaging (fMRI) is an imaging technique that indirectly measures brain activity by detecting blood oxygenation level dynamics that are believed to be the result of active neuronal communities.

Since the early 1990's, fMRI has become one of most popular and powerful imaging techniques. fMRI has been utilized by several different fields, such as neuroscience, psychology, sociology, and medicine, and has been innovative in helping clinicians and researchers in diagnosing pathologies and understanding brain function in animals and humans.

The workshop will focus on the basics involved in fMRI studies and include explanations regarding: 1) the biological response fMRI is based upon 2) basic components and physics of fMRI and equipment 3) fMRI designs and studies 4) fMRI data analysis 5) and fMRI study examples.



Presentation Slides (PowerPoint)

Background Reading:

www.fmrib.ox.ac.uk/education/fmri/introduction-to-fmri
www.arts.uwaterloo.ca/~jdancker/fMRI/fMRI_OL.htm
http://psychology.uwo.ca/fmri4newbies/
http://www.indiana.edu/~panlab/fmriDocs/studyDesign.pdf


Software:

SPM
AFNI

For more examples and more information, see the QuaSSI Backgrounder on functional MRI


A light lunch will be provided.  


Spring 2007 QuaSSI Seminar Schedule
Wednesday unless otherwise indicated. Specific titles are subject to change.
Date/Time/Location Seminar Details  
March 23
9:30-3:30 p.m.
Pond 302
Clark Glymour
Richard Scheines
Department of Philosophy
Center for Automated Learning and Discovery
“Causal Models and Statistical Data”

A great many applications of statistics in the natural and social sciences aim at finding or confirming causal hypotheses. While causal inferences from observational data have been the practical basis for a great deal of quantiative social science, the reliability of such inferences have been dismissed by theorists without much study. For fifteen years, Richard Scheines, Clark Glymour, and Peter Spirtes (aka "The TETRAD Group") have worked together and with students on the representation of causal hypotheses by "Bayes nets," on methods for finding such representation from data, and on the reliability limits of any possible methods.

Work from the TETRAD group has appeared in the books Discovering Causal Structure (1987), Causation, Prediction, and Search (2nd ed., 2001), Causation, Computation, and Discovery (1999), and The Mind's Arrow: Bayes Nets and Graphical Causal Models in Psychology (2001) and in journals of a wide range of disciplines, including philosophy, statistics, computer science, economics, psychology, sociologly, and medicine.

Schedule:
9:30 - 11:15 Pond 302 Introduction to the TETRAD Project: Causal Models & Statistical Data
11:30 - 1:00 Pond 302 Open Issues in Reliable Causal Inference
1:00 - 2:00 Pond 302 *Lunch
2:00 - 3:30 Pond 302 *Workshop: TETRAD Software for Causal Model Discovery
*Registration required (lunch and workrshop only). Email kjoyce@psu.edu by Wednesday, March 21.

Background Reading:


Software:
TETRAD: A program for creating, simulating data from, estimating, testing, predicting with, and searching for causal/statistical models.
 
February 21
12:10-1:30 p.m.
Pond 302
David Hunter
Department of Statistics
Center for Infectious Disease Dynamics
“Exponential Random Graph Models for Social Networks”

Consider a case in which we observe, for a sample of individuals, demographic characteristics as well as the presence of a particular type of relationship (such as friendship) between pairs of individuals. Suppose we wish to determine how the individuals' characteristics predict the presence or absence of relationships based on the observed sample. If we assume that every potential relationship is independent of every other, then logistic regression will suffice. However, one of the tenets of social network analysis is that social structure itself drives the formation of networks, meaning that relationships are not independent. This talk addresses statistical models for dealing with this situation, along with the special computational techniques that must be employed due to the mathematical intractability of the estimation problem. We illustrate using data on friendships among high school students.

Background Reading:
Inference in Curved Exponential Family Models for Networks (pdf)
Curved Exponential Family Models for Social Networks (pdf)
Goodness of Fit of Social Networks (pdf)

Software:
Statnet: R package for the representation, visualization, analysis and simulation of social network data.

A light lunch will be provided.  



Fall 2006 QuaSSI Seminar Schedule
Monday unless otherwise indicated. Specific titles are subject to change.
Date/Time/Location Seminar Details  
October 16
12-1:30 p.m.
Pond 302
Stephanie Lanza
Methodology Center / Health & Human Development
“An Introduction to Latent Class Analysis & Several Extensions”

This talk provides a conceptual introduction to latent class analysis (LCA) and its extension to multiple-groups LCA and LCA with covariates. Model specification will be discussed, and the approach will be demonstrated using an empirical example on heavy drinking behavior. New SAS software for latent class analysis will be demonstrated.

Presentation Slides (pdf)

Latent class and related mixture models have application across the social sciences, with examples being found in public health, psychology, political science, economics, sociology, and linguistics, among others. For more, read the QuaSSI Backgrounder on Latent Class Analysis.

A light lunch will be provided.  
October 23
12-1:30 p.m.
Pond 302
Kevin Murphy
Department of Psychology
“Power Analysis for Traditional and Modern Hypothesis Tests”

The talk lays out the logic and application of power analysis, and describes a very general model for such analyses that is based on the noncentral F distribution. The application of power analysis to alternates to the conventional null hypothesis, in particular the hypothesis that treatments have a nontrivial effect, are described. Prof. Murphy is the author, with Brett Myors, of Statistical Power Analysis: A Simple and General Model for Traditional and Modern Hypothesis Tests (2004, Erlbaum). This book can be perused online by the Penn State community via netlibrary.

Presentation Slides (PowerPoint)

Background Reading:
Murphy and Myors, 1999 (pdf)
Pollard & Richardson on Type I Errors (pdf)

Power analysis has application across the social sciences. For examples and more information, see the QuaSSI Backgrounder on Statistical Power Analysis.

A light lunch will be provided.
October 30
12-1:30 p.m.
Pond 302
Jee-Kwang Park
QuaSSI Postdoctoral Fellow
“Common Trend Models for Time Series Analysis”

Common trend models are time series analogues of factor analysis. Just as (static) factor analysis uncovers the latent factor(s) common to a set of variables, common trend models find unobserved trend(s) common to a group of related time series. The talk discusses three approaches to estimating common trend(s) among multiple series: Dynamic Factor Analysis, Time Series Factor Analysis, and the State-Space common trend model.

These models have potential application across the social sciences. They are common in economics, where common trend models are applied to time series of exchange rates, stock prices, macroeconomic indexes and so on. The talk will demonstrate two applications from political behavior and political economy, one analyzing the set of presidential approval time series produced by independent polling houses, and the other analyzing the set of tariff rates series produced by different countries.

Presentation Slides (PowerPoint)

For more examples and more information, see the QuaSSI Backgrounder on Common Trend Models

A light lunch will be provided.
 
November 2 (Thurs.)
2-5 p.m.
Pond 302
QuaSSI Workshop
Christopher Achen
Roger Williams Straus Professor of Social Sciences
Department of Politics, Princeton University
“The New Political Methodology”

Prof. Achen discusses the engagement between empirical and formal approaches to political science. He calls for more sophisticated approaches to data analysis, including the establishment of firm microfoundations in human behavior and avoiding the dangers of ``garbage can'' model specifications.

Recommended reading for this workshop:

Toward a New Political Methodology: Microfoundations and ART. (Perspectives on Politics, 2002)

Let's Put Garbage-Can Regressions and Garbage-Can Probits Where They Belong. (Conflict Management and Peace Science, 2005)


Further applications can be found in:

Christopher H. Achen. Expressive Bayesian Voters, their Turnout Decisions, and Double Probit: Empirical Implications of a Theoretical Model. (This paper will be presented in a Department of Political Science Seminar on Friday, Nov.3, in Pond 302.

Christopher H. Achen and Larry M. Bartels. Blind Retrospection: Electoral Responses to Drought, Flu, and Shark Attacks.


Further background on the EITM movement in political science can be found in:

Jim Granato and Frank Scioli. Puzzles, Proverbs, and Omega Matrices: The Scientific and Social Significance of Empirical Implications of Theoretical Models (EITM). (Perspectives on Politics, 2004)


Further discussion of parsimony in models of international relations can be found in other entries in the exchange published in the journal Conflict Management and Peace Science:

James Lee Ray. Explaining Interstate Combflict and War: What Should Be Controlled For?. (2005).

John R. Oneal and Bruce Russett. Rule of Three, Let It Be? When More Really is Better.. (2005).

James Lee Ray. Constructing Multivariate Analyses (of Dangerous Dyads). (2005).

November 13
12-1:30 p.m.
Pond 302
Donna Peuquet
Department of Geography and The GeoVista Center
“Time for Change: An Integrated Environment for Representation and Analysis of Complex Space-Time Processes”

Because progress in geographic information systems (GIS) technology has historically relied on a fragmented gathering of approaches inherited from cartography, imposed by hardware, or borrowed from other computer-related fields, we are faced with the current situation in which (1) GIS still live in a static world, focusing on the spatial, but ignoring the temporal dimension, and (2) the initial promise of true analytical capability within an integrated environment that truly combines the computational power of computers with the perceptual powers of the human user has still not been fully realized almost 40 years after the first GIS became operational.

While filtering through vast amounts of data now available to find patterns and associations requires new analytical techniques, it also requires new ways to represent the data. Effective use of heterogeneous and multidimensional data collections for undirected investigative discovery, as well as more directed analyses, requires an integrated approach for organizing evidence in a way that facilitates the task of the human user. Such an integrated approach must combine a multi-representation database strategy with diverse but coordinated visualization strategies within a unified conceptual scheme so that large volumes of complex information can be analyzed and synthesized.

In the talk, I will first describe some current issues relating to space-time representation on a conceptual level, and then introduce the approach used in a prototype integrated space-time database-visualization system currently under development relating to those issues.

Dr. Peuquet is a Professor of Geography at Penn State. Current research interests include theory of geographic knowledge representation, spatio-temporal data models, spatial cognition, spatial analysis methodologies, geographic database design, and the use of GIS in epidemiological research. Her recently published book, entitled Representations of Space and Time, with support in part from a Guggenheim Fellowship, develops an integrated perspective on philosophical, cognitive and technical issues on spatial and space-time representation.

A light lunch will be provided.
November 27
New Date!
12-1:30 p.m.
Pond 302
Aleksandra Slavkovic
Department of Statistics
“Statistical Methods for Data Privacy, Confidentiality and Disclosure Limitation”

Federal statistical agencies and non-government survey organizations collect data on individuals and groups while striving to maintain confidentiality. These data are often summarized in tables of counts (i.e., contingency tables) or tables of rates (i.e. conditional observed frequencies), and then released to various agencies, researchers, and policy makers for analysis that could have influence on policy decisions.

Statistical disclosure limitation (SDL) applies statistical tools to the problem of limiting releases of sensitive information about individuals and groups that are part of statistical databases while allowing for proper statistical inference. This talk gives an overview of issues and SDL approaches relevant for addressing the questions: (1) What social science data are releasable from a table with small counts that will not raise confidentiality concerns? and (2) Will the released data be useful for statistical inference?

A light lunch will be provided.
December 4
CANCELLED
Brian Smith
College of Information Sciences and Technology / College of Education
“Analyzing Competitive Resource Allocation and Cultures of Learning in Fantasy Sports League Data”

This seminar will be rescheduled for the Spring 2007 semester.
December 11
12-1:30 p.m.
Pond 302
 
Sung Jae Jun
Department of Economics
“Efficient Multivariate Quantile Regression Estimation”
(joint work with Joris Pinske)

Just as the mean provides only a limited picture of any distribution, conditional mean estimation with OLS regression provides only a limited picture of relationships. Quantile regression is focused on providing estimates of conditional quantile relationships, such as how compression in percentiles of the wage distribution varies for workers of low and high education. An excellent and gentle introduction -- with motivation, illustrative examples, and discussion of software implementations -- was published in 2001 in the Journal of Economic Perspectives by Koenker and Hallock and is recommended as background reading for those unfamiliar with the quantile regression (Link through JSTOR -- requires PSU IP address or VPN.)

In this seminar, Prof. Jun presents research with Joris Pinske on a proposed efficient semiparametric estimator for the multivariate linear quantile regression model. The paper can be downloaded here.

A light lunch will be provided.
 
 

Past QuaSSI Seminars

Spring 2006


Fall 2005


  burtmonroe@psu.edu