2004-2005 ITV Course Schedule
SPRING I:
"Spatial Data Modeling" - David Darmofal
Fridays: 11:00-1:00 CST
Schedule: Jan. 21, 28, Feb. 4, 11, 25, Mar. 4, 11
Email: darmofal.3@osu.edu
"Event History" - Jan Box-Steffensmeier
Weds: 1:30-3:30 CST
Schedule: Jan. 19, 26, Feb. 2, 9, 23, Mar. 2, 9
Email: steffensmeier.2@osu.edu
Course Descriptions
Spatial Data Modeling
Instructor: David Darmofal, Senior PRISM Fellow at Ohio State University
Email: darmofal.3@osu.edu
Time: 11:00am - 1:00pm CST, Fridays
Spring I Schedule:
January 21, 28
February 4, 11 (NO CLASS February 18), 25
March 4, 11
Description:
This course introduces students to modeling techniques for spatial data. Many of our models in political science (e.g., of policy diffusion, interstate conflict, and political communication) posit that spatial context shapes political behavior: the behavior of observed units is influenced by units in close proximity. Recent advances in spatial econometrics combine with the growing availability of geo-coded data to allow for much more rigorous testing of spatial influences on political behavior than has previously been possible.
Spatial econometrics remain an important tool for political scientists even when behavior is purely atomistic, and is not shaped by social context. Atomistic neighbors' shared values on omitted variables can induce spatially autocorrelated errors, producing inefficient OLS estimates, biased standard errors, and erroneous inferences on covariates of substantive interest. Spatial econometric models are able to correct for these spatially autocorrelated errors.
The course will begin by examining the problems that spatial autocorrelation (a systematic relationship between values on a variable of interest and geographic location) poses for standard approaches to statistical inference. Next, we will consider global tests of spatial autocorrelation, as well as local indicators of spatial association (LISA) statistics, such as Moran's I and Geary's c, which identify which observations are spatially autocorrelated. This portion of the course will also introduce students to geographic information systems (GIS) software, which is useful for many applications in political science.
Next, we will turn to strategies for modeling spatial dependence. Here, we will begin by considering the familiar OLS regression model as well as a set of spatial diagnostics that measure whether spatial autocorrelation persists in the presence of covariates in the OLS specification. Next, we will consider spatial econometric alternatives for modeling this spatial dependence. Here, we will focus on two principal classes of spatial regression models. Spatial lag models are applicable when spatial context shapes political behavior. Spatial lag models incorporate contextual influences via a spatially lagged dependent variable. Spatial error models are applicable for aspatial processes that are influenced by omitted variables. Here, the spatial dependence is modeled via a spatially lagged error term. We will conclude the course by considering emerging frontiers in spatial econometrics, such as spatial models for limited dependent variables, space-time models, and Bayesian hierarchical models.
Students are expected to have previous training in probability theory and regression analysis. Students should also bring to the class some interesting questions about spatial influences on political behavior. Students will be encouraged to work with geo-coded data sets of their own choosing that allow for empirical analyses of spatial effects on political behavior.
Event History
Instructor: Jan Box-Steffensmeier, Ohio State University
Email: steffensmeier.2@osu.edu
Times: 1:30pm - 3:30pm CST, Wednesdays
Spring I Schedule:
January 19, 26
February 2, 9 (NO CLASS February 16),23
March 2, 9
Description:
Social science theories are increasingly focused on change processes and temporal data are becoming widely available. Event history methods are ideal for studying temporal change. They address not only whether an event occurred, but when the event occurred. For many research questions in social science, the timing or history of social change is at least as interesting as understanding the culminating event. Research designs incorporating "history" into the analysis promises greater analytical leverage than other designs.
Event history analysis is longitudinal and involves the statistical examination of longitudinal data collected on a set of observations. While a wide variety of statistical models may be constructed for event history data, at the most basic level, all event history models have some common features. The dependent variable measures the duration of time that units spend in a state before experiencing some event. Generally, a researcher knows when the observations enter the process, i.e., when the history begins, and when, and whether or not, the process ends (with the occurrence or nonoccurrence of some event). Analysts are typically interested in the relationship between the length of the observed duration and independent variables, or covariates, of theoretical interest. A statistical model can then be constructed to link the dependent variable to the covariates. Inferences can be made regarding the influence of the covariates on the length of the duration and the occurrence (or nonoccurrence) of some event.
These methods have many advantages and allow new questions to be addressed. Event history data are becoming more available in all areas of empirically oriented political science. Applications include the duration of peace, the duration of unemployment, the length of time a cabinet is in place, when a challenger enters a congressional race, the duration of congressional careers, when a policy is likely to be adopted by the states, or how long it takes to complete a dissertation. The course will thoroughly describe different models for different kinds of duration data, document the assumptions underlying these different models, and consider goodness-of-fit indices and diagnostic techniques, i.e., residual and specification analysis.
Back to Courses Page.
