Empirical Models for Time-Series, Cross-Section Data

Instructor(s):

• Robert J. Franzese, Political Science, University of Michigan

Time-series cross-section (TSCS) data harness both cross-temporal and cross-spatial variation to maximize empirical leverage for theory evaluation. However, this powerful data structure also requires careful consideration of temporal and spatial (cross-unit) heterogeneity, temporal and spatial dynamic processes, and potentially complex stochastic error structures. This course covers specification, estimation, interpretation, and presentation of empirical models that are appropriate for TSCS data. The workshop begins by discussing the nature of pooled data and the ways that they deviate from the assumptions associated with the classic linear regression model. The course then addresses a number of issues that are typically associated with TSCS data: fixed or stochastic unit-heterogeneity, complex error structures, and temporal/spatial correlation and dynamics. We consider a variety of methodological strategies for confronting these issues in an effective manner, such as: fixed or random-effect models and associated tests; feasible-generalized-least-squares (FGLS); consistent coefficient-estimate variance-covariance (HAC) estimators; and temporal and spatial-lag models. The course concludes with a brief overview of TSCS models for noncontinuous dependent variables.

Note: Participants in this workshop should have a solid understanding of the classical multiple regression model in matrix form.

Dates:  May 18-22 

Fee:  Member: $1800; Non-member: $4000

This course is limited to 20 participants.