2013 Methods Workshop: Analysis of Panel Data - Random and Fixed Effect Modeling
Dr.
Günther Fink (
gfink@hsph.harvard.edu) led a workshop from 2-5:30 pm on
Jan. 23, 2013 in Oslo, Norway.
The
goals of the methods workshop were to:
- Introduce participants to longitudinal data and the main advantages of panel data
- Explore the differences between random effects, multilevel and fixed effects models, with a particular focus on the underlying statistical assumptions and their implications for causal inference
Topics
covered included:
- Definition and structure of longitudinal data
- Conceptual advantages of longitudinal data compared to cross-sectional data
- Identification of statistical models with panel data: "between" and "within" variation
- Empirical challenges: correlation of residuals within groups; lack of temporal variation; attrition and loss to follow-up; and non-random temporal (within) variations and instrumental variable estimation as potential solution
READ MORE about the workshop >>
2012 Methods Workshop: Analysis for Causal Inference With Cross-Sectional Data
T. Paul Schultz of Yale University (paul.schultz@yale.edu) led a workshop for researchers from 2-5:30 pm EDT on Jan. 18.
The goals of the methods workshop were to:
- Introduce workshop participants to some of the limitations of correlation-based analyses in empirical studies.
- Highlight the potential advantages of trying to infer causal relationships through the use of non-experimental techniques (e.g. instrumental
variables, control functions, matching, discontinuities) for cross-sectional data.
Topics covered:
- Causal inference in non-experimental studies using instrumental variables, and other identification methods.
- The advantages of individual and family panel data and the limitations of attrition.
- The advantages and disadvantages of reduced form versus structural estimation of causal models and drawing relevant policy inferences.
- Specification, estimation, and inference in models that may include individual, community, and group effects.
- The advantages and disadvantages of experimental versus non-experimental methods in inferring causality.
READ MORE about the workshop >>