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 >>