Panel data errors are almost always serially correlated and heteroskedastic. Stata offers:
Most interesting finding: With few clusters (<20), cluster-robust SEs are too small. Use clustse (user-written) or bootstrap with vce(boot, cluster(id)).
xtdescribe // Summary of panel structure: balanced? gaps?
xtsum // Summary statistics within and between panels
xttab indvar // Tabulation by panel ID
xtline yvar // Line plots for each panel (use as check for outliers)
Example:
xtsum hours wage
This output is invaluable: it shows overall variation, between variation (across individuals), and within variation (over time for the same individual). High within variation is good for fixed effects models.
Panel data errors are rarely i.i.d. They typically exhibit heteroskedasticity and serial correlation (within unit). Use cluster-robust standard errors.