nr year school exper union
1 13 1980 14 1 no
2 13 1981 14 2 yes
3 13 1982 14 3 no
4 13 1983 14 4 no
5 13 1984 14 5 no
6 13 1985 14 6 no
7 13 1986 14 7 no
8 13 1987 14 8 no
9 17 1980 13 4 no
10 17 1981 13 5 no
11 17 1982 13 6 no
12 17 1983 13 7 no
13 17 1984 13 8 no
14 17 1985 13 9 no
15 17 1986 13 10 no
16 17 1987 13 11 noPanel Data Introduction
Tobias RÜttenauer
This course
Outline
- Introduction Panel Data
- Variance Components
- Estimators: FE & RE & Diff-in-Diff
- Dynamic Diff-in-Diff
- Fixed Effects Individual Slopes
Me
Lecturer in Quantitative Social Science at UCL
Environmental Sociology
Further Materials
This material in Handbook format: Rüttenauer and Kapelle (2024)
Slides by Josef Brüderl and Volker Ludwig. See also Brüderl and Ludwig (2015).
Books:
Intuitive: Allison (2009)
Comprehensive and formal: Wooldridge (2010)
For R experts: Croissant and Millo (2019)
General introductions to causal estimation techniques: Angrist and Pischke (2015), Cunningham (2021), Firebaugh (2008), Huntington-Klein (2021)
The books by Cunningham (2021) (Link) and Huntington-Klein (2021) (Link) are freely available online!
Why panel data analysis?
In empirical social sciences, we are often interested in the causal research questions: we want to investigate questions of cause and effect.
However, randomized controlled trials (RCT) are often infeasible (e.g. effects of education, marriage, pregnancy).
A potential middle ground between those two approaches: “compare alike with alike”.
With panel data, we observe the same unit (person, region, or country) repeatedly over time. We can then not only compare two different units to each other. We can also compare a unit in an earlier stage to the same unit in a later stage.
Panel Data Structure
Usually cross-sectional data is organized as a matrix, where rows represent the observation / individual and the columns hold the variables. In panel data settings, we need to add the dimension of time. There are two ways to do so:
Long format: \(N \times T\) observations (rows), with variables “id” and “time”.
Wide format: \(N\) observations, and \(T \times K\) variables, which one variable for each time-period.
Panel Data Structure
Let’s have a look at the “Males” data of the plm package.
Panel Data Structure
Moreover, there two types of panel data:
Balanced: Contains information for each unit at each time period
Unbalanced: Some units have missing information at some time periods
Panel Attrition
Panel surveys
Some examples
The Comparative Panel File provides an infrastructure of data harmonisation across various panels. See Turek, Kalmijn, and Leopold (2021).
Panel transformation
A nice feature of panel data is that we can do some within-person transformation. For instance we can calculate the lags and leads, or the first differences of data.
Note
always make sure the data is sorted properly before you do!
Panel transformation
Person-specific summary values
Panel transformation
Lag and first difference
# Order data
Males <- Males[order(Males$nr, Males$year),]
# Lag (last years value)
Males$lag_wage <- ave(Males$wage,
Males$nr,
FUN = function(x) dplyr::lag(x, n = 1))
# First difference (this years value minus last years value)
Males$fd_wage <- ave(Males$wage,
Males$nr,
FUN = function(x) x - dplyr::lag(x, n = 1))Panel transformation
nr year wage m_wage lag_wage fd_wage
1 13 1980 1.1975402 1.255652 NA NA
2 13 1981 1.8530600 1.255652 1.1975402 0.65551979
3 13 1982 1.3444617 1.255652 1.8530600 -0.50859832
4 13 1983 1.4332133 1.255652 1.3444617 0.08875166
5 13 1984 1.5681251 1.255652 1.4332133 0.13491174
6 13 1985 1.6998909 1.255652 1.5681251 0.13176586
7 13 1986 -0.7202626 1.255652 1.6998909 -2.42015352
8 13 1987 1.6691879 1.255652 -0.7202626 2.38945049
9 17 1980 1.6759624 1.637786 NA NA
10 17 1981 1.5183982 1.637786 1.6759624 -0.15756420
11 17 1982 1.5591905 1.637786 1.5183982 0.04079228
12 17 1983 1.7254101 1.637786 1.5591905 0.16621961
13 17 1984 1.6220223 1.637786 1.7254101 -0.10338777
14 17 1985 1.6085883 1.637786 1.6220223 -0.01343405
15 17 1986 1.5723854 1.637786 1.6085883 -0.03620286
16 17 1987 1.8203339 1.637786 1.5723854 0.24794844References
<Home>
Allison, Paul David. 2009. Fixed Effects Regression Models. Vol. 160. Quantitative Applications in the Social Sciences. Los Angeles: Sage.
Angrist, Joshua David, and Jörn-Steffen Pischke. 2015. Mastering ’Metrics: The Path from Cause to Effect. Princeton: Princeton Univ. Press.
Brüderl, Josef, and Volker Ludwig. 2015. “Fixed-Effects Panel Regression.” In The Sage Handbook of Regression Analysis and Causal Inference, edited by Henning Best and Christof Wolf, 327–57. Los Angeles: Sage.
Croissant, Yves, and Giovanni Millo. 2019. Panel Data Econometrics with R. Hoboken, NJ: John Wiley and Sons.
Cunningham, Scott. 2021. Causal Inference: The Mixtape. New Haven and London: Yale University Press.
Firebaugh, Glenn. 2008. Seven Rules for Social Research. Princeton, N.J. and Woodstock: Princeton University Press.
Huntington-Klein, Nick. 2021. The Effect: An Introduction to Research Design and Causality. Boca Raton: Chapman & Hall/CRC.
Rüttenauer, Tobias, and Nicole Kapelle. 2024. “Panel Data Analysis.” https://doi.org/10.31235/osf.io/3mfzq.
Turek, Konrad, Matthijs Kalmijn, and Thomas Leopold. 2021. “The Comparative Panel File: Harmonized Household Panel Surveys from Seven Countries.” European Sociological Review 37 (3): 505–23. https://doi.org/10.1093/esr/jcab006.
Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and Panel Data. Cambridge, Mass.: MIT Press.