Individual Fixed Effects & Panel Data Estimators Flashcards
Advantages and assumptions of individual fixed effects
The ideal counterfactual may be the individual himself in the previous period
If there is variation in the treatment over time, you can compare the same individual with and without treatment
Requires longitudinal/panel data: observations for the same individual at different points in time
Individual fixed effects estimation uses within-individual variation
Limitations of individual fixed effect estimation
- You only use within-individual variation in T, so only estimate effect of time-varying treatment variables and not time-invariant treatment variables (that stay the same over time)
- As with twin fixed effects, measurement error problems are bigger when looking at within-individual changes
- Where is the with-in variation in T coming from? There can be time-varying shocks that influence both T and Y
Why does Granger causality not necessarily imply causality?
Correlation between T and Y can be due to:
1. T causes Y
2. Y causes T
3. Omitted variables affect both T and Y
Granger causality can distinguish between 1 and 2 by looking at the history of T and Y. But it cannot distinguish between 1/2 and 3. So if T Granger-causes Y, does not imply causal effect because could also be due to omitted variables.
Difference between event study and individual fixed effects
Event studies: all individuals in the panel get treatment, but at different times. Allows for anticipation before treatment and dynamic treatment effects after (graph).
Include individual fixed effects if you believe time-invariant unobserved factors play a big role.
Individual fixed effects assumes that the treatment effect is constant over time, whereas event studies allow the treatment effect to vary over time
Why is panel data not a sufficient condition for causality?
Likely that within-variation is not random, but due to unobserved factors / omitted variables
