2003-2004 ITV Course Schedule

FALL I:

Charles Franklin -- "Bayesian Methods"
Weds. 1:30-3:30 CST
Sept. 10, 17, 24, Oct. 1, 8, 15, 22
Email: franklin@polisci.wisc.edu

FALL II:

Charles Franklin -- "Bayesian Methods II"
Weds. 1:30-3:30 CST
Oct 29, Nov. 5, 12, 19, 26, Dec. 3, 10
Email: franklin@polisci.wisc.edu

Herb Weisberg -- "Scaling and Dimensional Analysis"
Fridays, 11:00-1:00 CST
Oct 17, 24, 31, Nov. 7, 14, 21, Dec. 5
Email: weisberg.1@osu.edu

SPRING I:

Wendy Tam Cho -- "Statistical Computing"
Wednesdays, 1:30-3:30 CST
Jan. 21, 28, Feb. 4, 11, 18, 25, Mar. 3
E-mail: wendy@cho.pol.uiuc.edu

Jan Box-Steffensmeier, John Freeman, and Jon Pevehouse -- "Time Series Analysis"
Fridays 11:00-1:00 CST
Jan. 23, 30, Feb. 6, 13, 20, 27, Mar. 5
Email: steffensmeier.2@osu.edu, freeman@polisci.umn.edu, pevehous@polisci.wisc.edu

SPRING II:

Brian Gaines -- "Panel Data Analysis"
1:30-3:30 CST
Mar. 10, 31, Apr. 7, 14, 21, 28, May 5
Email: bjgaines@uiuc.edu

Jan box-Steffensmeier, John Freeman, and Jon Pevehouse -- "Time Series Analysis II"
Fridays, 11:00-1:00 CST
Mar. 12, Apr. 2, 9, 23, 30, May 7
Email: steffensmeier.2@osu.edu, freeman@polisci.umn.edu, pevehous@polisci.wisc.edu

2003-2004 Course Descriptions

Bayesian Methods II

Instructor: Charles Franklin, University of Wisconsin
Email: franklin@polisci.wisc.edu
Times: 1:30 - 3:30 CST, Wednesdays
Fall I Schedule:
September 10, 17, 24
October 1, 8, 15, 22
Fall II Schedule:
October 29
November 5, 12, 19, 26
December 3, 10

Description:

This course introduces Bayesian methods for data analysis in the social sciences. Bayesian methods provide a flexible and powerful approach to complex statistical models and have a theoretical elegance and clarity that is impressive. Bayesian models inherently recognize and incorporate subjective judgments of the researcher, which is the source of both their great power the controversy surrounding their use. We will discuss some of the epistemological issues raised by Bayesian methods as well as their application. We will cover the basic concepts of Bayesian statistical inference. There are several sets of tools needed to do applied Bayesian modeling. First we need to review some probability theory. Second, we will develop the fundamental notion of Bayes theorem as a foundation for statistical inference. We'll also see how likelihood is incorporated within the Bayesian framework. Finally, we'll explore the world of applied Bayesian modeling. This will include learning some new software tools using WinBugs and S-Plus/R. The main focus of the course will be application of Bayesian models to cutting edge issues in social science modeling. This will include a number of applied readings. We'll take time to discuss these applications in class in order to develop a feel for what research that takes a Bayesian approach "feels" like.

The goal of the class is to develop the necessary theoretical understanding to correctly apply Bayesian models using modern software. A second goal is to reach a level of understanding that will support further reading and learning on your own. The syllabus points the way to further reading throughout in the "Advanced Topics" sections.

Texts:

Andrew Gelman, John B. Carlin, Hal S. Stern and Donald B. Rubin. 1995. Bayesian Data Analysis . New York: Chapman & Hall.
Jeff Gill. 2002. Bayesian Methods for the Social and Behavioral Sciences . New York: Chapman & Hall/CRC.

Scaling and Dimensional Analysis in Political Science

Instructor: Herb Weisberg, The Ohio State University
Email: weisberg.1@osu.edu
Course Webpage: http://www.class.osu.edu
Times: 11:00 - 1:00 CST, Fridays
Fall II Schedule:
October 17, 24, 31
November 7, 14, 21, 28
December 5

Description:

Dimensional perspectives are common in thinking about politics. This course covers the methodology of dimensional analysis -- the scaling techniques and their philosophical implications. The dimensional approach is based in geometry, so visual displays will be emphasized. Theoretical perspectives on methods, scaling, and dimensions will also be presented. The methods presented are a variety of techniques for scaling, broadly defined, including unfolding analysis, proximity scaling, Guttman scaling, cluster analysis, factor analysis, and multidimensional scaling. These methods provide means for:

data reduction (reducing a large number of variables into a smaller set of composites)
examining dimensionality (representing the data in terms of the smallest possible number of unobserved underlying factors)
measurement (scoring cases on the underlying dimensions and using those scores in further analysis)

The course will be presented as a "module" is part of an on-going program of graduate Interactive Television Program (ITV) in Advanced Political Methodology, a cooperative program by the Departments of Political Science on four CIC campuses (Illinois, Michigan, Ohio State, and Wisconsin) using interactive video and WebCT.

Requirements:

Students will be expected to use relevant computer programs and interpret the results.
A report (5 pages) on an article in your substantive field that uses the techniques treated in this class. Due: November 14.
A term paper analyzing a set of data of your choosing from a dimensional perspective, using whichever techniques are most appropriate. The best papers are generally those which use multiple dimensional techniques on the same data and compare the results. Due: Thursday December 4.
A takehome final exam will be given. Due: Wednesday December 10 at noon.

Required books:
No single textbook covers all of the topics in this course. The full-length books which have been written on these topics are generally out-of-print, partially out-of-date, very expensive, and/or overly mathematical. As a result, several paperbacks will be used in this course as text-substitutes:

Abbott, Flatland (or Burger, Sphereland)

Several Sage monographs on scaling:

Jacoby, Data Theory and Dimensional Analysis (Sage #78)
McIver & Carmines, Unidimensional Scaling (#24)
Kruskal & Wish, Multidimensional Scaling (#11)
Kim & Mueller, Introduction to Factor Analysis (#13)
Kim & Mueller, Factor Analysis (#14)
Long, Confirmatory Factor Analysis (#33)
Aldenderfer & Blashfield, Cluster Analysis (#44)

Statistical Computing

Instructor: Wendy Cho, University of Illinois
Email: wendy@cho.pol.uiuc.edu
Course webpage: http://cho.pol.uiuc.edu/~wendy/
Times: 1:30-3:30 CST, Wednesdays
Spring I Schedule:
January 21, 28
February 4, 11, 18, 25
Mar 3

Description:

This course will focus on aspects of statistical computing. Modern statistical packages provide many tools for analyzing data. One purpose of this course is to take a behind-the-scene look at some of these tools in an effort to move away from a simple point-and-click environment to a better understanding of some of these powerful tools. In addition, we will examine how advances in computing have led to advances in statistical analysis through tools such as simulation. A variety of statistical methods will be discussed in the context of specific data analysis problems. Topics may include distributions and random data, numerical methods, Monte Carlo experiments, and optimization (via e.g., maximum likelihood).

Although there are no formal prerequisites for this course, students should be comfortable using computers and should be conversant with some statistical software package. Although I prefer to do my work in a UNIX environment, you are welcome to work in either a UNIX environment or a windows environment. This course will emphasize the S-plus statistical package primarily through its free counterpart, R. Although I will not spend much time discussing implementation in other statistical package, you should feel free to use and complete assignments using any package that is appropriate for the tasks.

This syllabus as well as some class notes, announcements, and problem sets will be available at the web site http://cho.pol.uiuc.edu/ps493/ . This web site will be updated regularly as the course progresses. Data sets for the computer assignments and various other items of interest will also be available at the class web site.

Time Series Analysis

Instructors:
Jan Box-Steffensmeier, The Ohio State University
email: steffensmeier.2@osu.edu
John Freeman, University of Minnesota
email: freeman@polisci.umn.edu
Jon Pevehouse, University of Wisconsin
email: pevehous@polisci.wisc.edu
Syllabus: Download the syllabus
Course Webpage: View Webpage
Times: 11:00-1:00 CST, Fridays
Spring I Schedule:
January 23, 30
February 6, 13. 20, 27
Mar 5
Spring II Schedule:
March 12
April 2, 9, 23, 30
May 7

Description:

This course considers statistical techniques to evaluate social processes occurring through time. The course introduces students to time series methods and to the applications of these methods in political science. After a brief review of the calculus of finite differences and other estimation techniques, we study stationary ARMA models. In the next section of the course, we examine a number of important topics in time series analysis including "reduced form" methods (granger causality and vector autogression), unit root tests, near-integration, fractional integration, cointegration, and error correction models. Time series regression is also discussed (including pooling cross-sectional and time series data). We learn not only how to construct these models but also how to use them in policy analysis.

We expect students to have a firm grounding in probability and regression analysis and to bring to the course some interesting questions about the dynamics of political processes. The emphasis throughout the course will be on application, rather than on statistical theory. However, the focus of most lectures will be statistical theory. Homework will revolve as much as possible around the time series you are interested in understanding. To that end, students will need to gather time serial data for analysis during the first week of class (this data need not be used throughout the term, though that would make your life easier). The length of the series should be at least 40 time points; longer series are better than shorter ones.

This is the first part of a fourteen-week seminar team-taught by Professors John Freeman, Janet Box-Steffensmeier, and Jon Pevehouse. Students are strongly encouraged to take both parts of the course.

Panel Data Analysis

Instructor: Brian Gaines, University of Illinois Urbana-Champaign
Email: bjgaines@uiuc.edu
Times: 1:30 - 3:30 CST, Wednesdays
Spring II Schedule:
March 10, 31
April 7, 14, 21, 28
May 5

Description:

This course will introduce the very large and rapidly growing statistical literature on analyzing data in which observations are a cross section of some units (e.g. individuals, countries, states, firms) over multiple time periods. I will use "panel data" in a broad sense to mean any data spanning multiple dimensions, usually (but not necessarily) time and space, but our focus will mostly be on cases in which we have much larger Ns than Ts. In general terms, the advantage to having time and space variance is that one can avoid inferring that inter-personal differences across units are equivalent to inter-temporal differences within units. For instance, in a cross-section if we find that age is a significant predictor of voting, we often infer that this result implies a forecast that all members of the relevant population will vote with increasing probability as they age. From a time series on just one unit, we might make the companion inference: that an observed life-cycle change in one unit implies across-age variation in a population that is heterogeneous in age in any given time period. With data that vary on both dimensions, one can better disentangle inter-unit and inter-temporal variance, and thus, in many contexts, adjudicate between rival behavioral theories. The increasing availability of large data sets and powerful computers has made panel models more and more prominent in econometrics.

None of the ITV classes assume any specialized skills or prerequisites beyond familiarity with classic regression and a degree of comfort with matrix calculations. In a perfect world, one might study both maximum likelihood estimation and time series analysis in advance of panel data models, but I will not assume that students have followed that particular sequence.

In seven weeks, we cannot possibly cover even a full introductory text, let alone the whole field, so I must emphasize the first word in the title: this course should be regarded as only a first step. We will work from Cheng Hsiao's Analysis of Panel Data (Cambridge, 2003), with the aim of covering less than half of the book. A few political science applications will be added, and we'll deal with computing issues at least briefly. However, the course will probably, given its brevity, emphasize theory at the expense of practice.

Wed 10/03: Introduction, Analysis of Covariance
Wed. 31/03: Variable Intercepts, Fixed Effects
Wed. 07/04: Variable Intercepts, Random Effects
Wed. 14/04: Heteroskedasticity, Serial Correlation
Wed. 21/04: Simultaneous Equations
Wed. 28/04: Guest Lecture: Douglas Rivers (Stanford)
Wed. 05/05: A brief survey of excluded topics: dynamic models, limited dependent variables, unbalanced panels, data fields, multi-level models, etc.

An additional caveat is that this will be the beta-if not alpha-version of this course, i.e. I have not taught it before. I'll determine what the precise evaluands will be later, but I anticipate that there will be a couple of problem sets, and that I'll have students make short presentations in some sessions.

Required text: Hsiao, Cheng. 2003. Analysis of Panel Data, 2nd. Cambridge: Cambridge University Press.

Recommended: Baltagi, Badi H. 2001. Econometric Analysis of Panel Data, 2nd Ed. Chichester/NY: John Wiley and Sons, Ltd.