Panel data & fixed effects Flashcards
What is panel data?
Uses cross-sectional and a time dimension… repeated observations on the same unit (across time)
What are the advantages of panel data?
1) Robustness - Controls for time-invariant unobservable effects (reduces omitted variable bias)
2) Efficiency - more data and variation (less collinearity)
What are the disadvantages of panel data?
1) Practical - Observations on the same individual are not independent and many panel data sets are incomplete
2) Fundamental - may not be representative over time and potentially changing sample composition
What do fixed effects do?
They absorb variability coming from time-invariant individual characteristics.
How do you implement fixed effects?
Include a dummy variable for each individual, which will absorb the time-invariant characteristics (e.g: gender). This will help to take out as much information as possible which explains Yit (taking out from Uit) and makes other estimated beta coefficients more precise and unbiased.
How does the first difference model compare to FE model?
When t=2, they are equivalent (but less efficient), however, when t>2, the strict exogenity assumption of FE has been violated and they have very different results.
What is the random effect model (or the between estimator)?
It is the weighted average of the between estimator and the within estimator.
How do you test if the coefficients of the FE and RE approaches are the same?
Use the Hausman test: Estimate the models for both and if the results are ‘very different, then the random effects estimator is inconsistent and the FE model is more appropriate.
What are the limitations of cross-sectional data?
1) Can not analyse behaviour over time
2) Doesn’t allow us to determine cause and effect
3) Snapshot is not guaranteed to be representative
4) Can’t take into consideration time-invariant unobserved heterogeneity
What are the limitations of time series data?
1) Can not analyse heterogenous behaviour across individuals
2) Time periods do not form random samples
3) Rarely know the shape of the distribution
4) Observations within each series are not independent of each other (autocorrelation)
How does including FEs impact the other beta coefficients?
It takes out of uit as much information as possible that explains it, therefore, making the beta coefficients more precise and unbiased
How do we implement the within estimator?
Apply OLS to equations in deviations from individual means, exploiting the within dimensions (variations across time) of the data
How do we apply the between estimator?
OLS applied to the equations in individual mean, exploiting the between dimensions (variations across individuals) from the data
What does the RE model require that is very unlikely?
That the regressors (xit) and the individual component (alphai) are uncorrelated
