This is the fundamental command for panel data.
* Basic syntax
xtset panel_id time_variable
| Feature | Pooled OLS | Fixed Effects (FE) | Random Effects (RE) |
| :--- | :--- | :--- | :--- |
| Command | reg y x | xtreg y x, fe | xtreg y x, re |
| Assumption | No individual effects | $\alpha_i$ correlated with $x$ | $\alpha_i$ NOT correlated with $x$ |
| Time-Invariant Vars? | Yes | No (Dropped) | Yes |
| Efficiency | N/A | Low | High |
| Best For | Preliminary analysis | Causal inference (observational) | Efficiency / Random sampling |
xtivreg wage (tenure = age) hours, fe
(First-stage: tenure instrumented by age.) stata panel data
For interpretation, compute marginal effects:
margins, dydx(experience) at(union=(0 1))
Raw panel data often arrives messy. Prepare it systematically. This is the fundamental command for panel data
When lagged dependent variable appears as regressor, FE is biased (Nickell bias). Use GMM estimators.
We model log wages (ln_wage) as a function of hours worked, age, and tenure. (First-stage: tenure instrumented by age
| Task | Command |
|------|---------|
| Declare panel | xtset id time |
| FE regression | xtreg y x1 x2, fe |
| RE regression | xtreg y x1 x2, re |
| Hausman test | hausman fe re |
| Cluster SE | , robust or vce(cluster id) |
| Lag variable | gen x_lag = L.x |
| Panel line plot | xtline y |
| Drop if no variation | xtpattern, gen(pat); drop if pat == "111111" |
| Fill gaps | tsfill, full |