46th GESIS Spring Seminar (2017)

Causal Inference with Observational Data

March 6-24, 2017, Cologne, Germany

The GESIS Spring Seminar (formerly ZA Spring Seminar) has been taking place in Cologne annually for more than 40 years. It offers three consecutive one-week courses in advanced methods of quantitative data analysis for Social Scientists, including a preparatory course in Mathematics for Social Scientists. Language of instruction is English. The Spring Seminar 2017 provides in depth knowledge of state of the art quantitative analysis techniques in the social sciences. It enables participants to apply state of the art quantitative analysis techniques to their own research projects and provides participants the opportunity to discuss their own research with interested colleagues and foremost experts.

March 6-10, 2017


Prof. Dr. Michael Windzio, Germany

Dr. Marco Giesselmann, Germany

Short Course Description:

Longitudinal data is widely discussed as an important means to validate causal interpretations. This course introduces the basic methods suitable to exploit this potential of panel data. We start with methods for categorical independent variables. Here, we introduce the simple Life Event Design (LED) and explain how this is related to the Difference-in-Difference Estimator (DiD). If the independent variable is measured on a metric scale, social scientists usually employ regression techniques, which is also the case for longitudinal data. Therefore, we discuss extensions of the simple regression framework addressing the properties and potentials of longitudinal data. Concretely, we introduce Fixed Effects (FE), First Differences (FD), and Hybrid Regression Models (HM) and discuss the differences and assumptions of these techniques. For research questions with categorical dependent variables, we introduce two applications of logistic regression suitable for the analysis of longitudinal data: the Conditional Logistic Regression, which resembles the benefits of FE, and techniques of Event History Analysis (EHA), which are particularly suitable if the researcher explicitly focuses transitions of the dependent variable. In all parts of the course, we put a strong emphasize on the intuitive understanding of the methods employed. All exercises are based on the data from the Socio Economic Panel Study (SOEP), which will be introduced during the course.

March 13-17, 2017


Prof. Kenneth A. Bollen, PhD, United States

Zachary Fisher, United States

Short Course Description:

This workshop is about Structural Equation Models (SEMs) and the statistical software to estimate such models. The course provides an overview of and experience in constructing and estimating SEMs. The topics treated include: path analysis, confirmatory factor analysis, simultaneous equation models, the incorporation of multiple indicators and measurement error into structural equations, alternative estimation procedures, and the assessment of model identification, fit, and modification.

March 20-24, 2017


Prof. Dr. Ben Jann, Switzerland

Dr. Rudolf Farys, Switzerland

Short Course Description:

Based on the potential outcomes notation of causal effects (a.k.a. the Rubin Causal Model) a variety of methods for causal inference from observational data have been developed or re-discovered over the last two decades and became increasingly popular in cutting-edge social science research. This course provides an introduction to these methods, explains their foundations and assumptions and discusses the conditions under which their use is appropriate. Topics covered are conceptual approaches such as the potential outcomes framework and directed acyclic graphs (DAG) as well as estimation methods such as matching, inverse probability weighting, instrumental variables, regression discontinuity, and difference-in-difference. Upon completion of this course, students should have acquired skill in the estimation, specification and diagnostics of the various methods and gained hands-on experience with those methods through the use of appropriate software and actual data sets. For the exercises in the computer lab, the course relies on Stata.