Panel data contain repeated observations for the same units. Typical data structures are individual data collected in two or more waves at the micro level or country data collected for several years at the macro level. This kind of data enable researchers to gain more leverage to test causal theories. However, these settings imply more complex error structures which invalidate ordinary least squares estimators and require models which take these structures into account. The course starts with a discussion of the assumptions underlying the regression model and the ways in which panel data violate these assumptions. It then proceeds to quick fixes that correct for these violations but do not exploit all the information contained in panel data. The next step is a focus on fixed and random effects models and attempts at bridging the two models, followed by models with varying slopes. We then turn to the time dimension and discuss methods to deal with dynamics, persistence in time, such as lagged dependent variables and estimators based on first differences.

Panel data contain repeated observations for the same units. Typical data structures are individual data collected in two or more waves at the micro level or country data collected for several years at the macro level. The great advantage of panel data is that they include information on the same units of analysis for several points in time. Researchers gain more leverage to test causal theories, because they can relate later states or changes to earlier states or changes and thus study the sequence of events within units. They can then test whether this sequence is generalizable by studying the variance across units. However, these settings imply more complex error structures which invalidate ordinary least squares estimators. In addition, the richness of the data can only be explored by means of models which take these structures into account. Such models do not summarize statistical associations across time and space but allow for heterogeneity. 

The course starts with a discussion of the standard Gauss-Markov assumptions underlying the regression model and the ways in which panel data violate these assumptions (day 1). It then proceeds to discuss quick fixes such as the Huber-White sandwich estimator or estimators of standard errors that take clustered data structures into account (day 2). These approaches may correct for standard violations of ordinary least squares but do not exploit all the information contained in panel data. Even worse, they may downweigh any information that suggests deviations from the single regression line summarizing the general statistical association or even raise trust in estimators that provide a strongly biased impression of the statistical associations.

The next step (day 3) is to abandon the assumption that a single line adequately represents the statistical association in the data and to control for heterogeneity across units, panel waves, or both. This is effectuated by fixed and random effects models, whereby the former calculates an intercept for each unit or panel wave, whereas the latter uses assumptions about the distribution of intercepts to provide a more parsimonious specification. We will also discuss attempts at bridging the two specifications by means of the Hausman-Taylor estimator. This discussion is followed by models which allow for even more heterogeneity by varying not only intercepts but also slope estimators (day 4). The seemingly unrelated regression model is introduced and various possibilities to explore conditional effects are discussed, as well as the relationship of this class of models with multilevel models. We will also touch upon the two-step approach to analysing panel data.

 Finally, we will turn to the time dimension and discuss methods to deal with dynamics and with persistence in time, such as lagged dependent variables and estimators based on first differences (day 5). We will introduce the difference-in-differences model, discuss nonstationarity and have a brief look at more advanced econometric approaches for dealing with the time dimension such as the Andersen-Hsiao model and the Arellano-Bond estimator.

The course is targeted at junior researchers who wish to explore panel data sets. It is practical and uses only the absolute minimum of mathematics and econometrics that is necessary for understanding what is going on behind the screen. 

In the lab exercises, several data sets, including individual-level and country-level data, will be used. Lab sessions will use the software package Stata. The course is set up as follows: First a topic is presented in the lecture, then participants are asked to work on a set of tasks in which they stepwise develop and test a model, using the ideas and formulae presented in the lecture. Finally, participants are encouraged to use the methods and procedures that are appropriate for their own research questions, which they can either bring along or develop in the course.

The course assumes proficiency with the multiple regression model and requires a basic understanding of matrix algebra.

Course Readings

Articles

Halaby, Charles N. (2004) Panel Models in Sociological Research: Theory into Practice, in: Annual Review of Sociology 30, 507-544.

Kittel, Bernhard & Hannes Winner (2005) How Reliable is Pooled Analysis in Political Economy? The Globalization-Welfare State Nexus Revisited, in: European Journal of Political Research 44, 269-293.

Rüttenauer, Tobias & Volker Ludwig (2023) Fixed Effects Individual Slopes: Accounting and Testing for Heterogeneous Effects in Panel Data or Other Multilevel Models, in: Sociological Methods & Research 52, 43-84.

Wawro, Gregory (2002) Estimating Dynamic Panel Data Models in Political Science, in: Political Analysis 10, 25-48.

Wilson, Sven E. & Daniel M. Butler (2007) A Lot More to Do: The Sensitivity of Time-Series Cross-Section Analysis to Alternative Specifications, in: Political Analysis 15, 101-123.

Books

Baltagi, Badi H. (2021) Econometric Analysis of Panel Data (6th edition), New York: Wiley.

Finkel, Steven E. (1995) Causal Analysis with Panel Data, Thousand Oaks: SAGE Univeristy Press.

Hsiao, Cheng (2003) Analysis of Panel Data, Cambridge: Cambridge University Press.

This description is subject to change at the discretion of the Instructor.

2 Credits

For completion of all work before and during the course, as outlined by the Instructor, and 90% participation and attendance of the course.

2 Additional Credits

Course specific extra assignments. These can include submitting assignments before the course, daily assignments, and/or a final assignment to be completed after the course as decided by the Instructor.  This description is subject to change at the discretion of the Instructor.