Topic 13: Panel Data Methods Flashcards
Do the unobserved factors that are correlated with the dependent variable change over time in fixed-effects estimation?
No, the unobserved variable in the error term (omitted variable) does not change over time.
Can you use fixed effects estimation to estimate causal effects despite having unobserved errors that are correlated with a dependent variable?
Yes, as long as the unobserved factors do not change over time.
What is the parallel trend assumption for difference in differencing?
That the change in the treated and control groups would have been the same if not for the treatment.
When does difference in difference estimators not work?
When some factor that changes over time affects the outcome for only the treated group.
Are first difference and fixed effects estimation computationally identical?
No, though they get the same estimates when T=2. When the errors are serially uncorrelated fixed effects is more efficient and the standard errors are valid.
In a model with two binary variables making three interaction terms, what is the difference in differences estimate, in terms of parameters?
The estimate will be the coefficient of interaction term of interest minus the coefficients of the other two interaction terms (it is already differenced from the fourth, base group).
Stata: How do you set up for a difference in difference estimation?
Use the “xtset id time” command to tell Stata the data is panel data.
Stata: How do you run a regression in first differences to remove ai from the composit error term?
After setting Stata to panel data, use “d.” before each variable to run as first differences.
Stata: How do you run a fixed effects regression with fixed effects to the id and year?
After setting Stata to panel data with the id and time variables, use “xtreg y x1 x2 x3…. , fe”, this will use fixed effects for the id. To add in the year fixed effects use a dummy for each year or use “i.year”
When comparing estimates from an OLS regression and a fixed effects regression, why might the OLS estimate be positively biased?
There may be a correlation between unobserved effects and the explanatory variables in each time period, causing positive omitted variable bias. The fixed effects estimator removes such fixed effects from the error term, removing the bias.
Why might you want to use a lagged version of a variable of interest in a fixed effects model?
You would want to include a lagged version in case the effect takes some time to show up.
Stata: how do you use first differences versus fixed effects regressions?
For first differences just set up for panel data, use “xtreg” and then “d.” in front of every variable. For fixed effects set up for panel data, use “xtreg”, and end with “, fe”
Why might a simple regression estimate be positive while the first differences and fixed effects estimators negative?
They differ because first difference and fixed effects removes the effect of a fixed part of the estimator, ai, and they are negative because the ai was positively correlated with x.
What is panel data?
Repeated cross sections, random cross-sectional samples at different time periods
What do difference in difference methods try to do?
Find a control group which wasn’t affected by treatment policy.
