Fixed and random effects panels
Distinguish within-firm change from stable firm differences
Panel models use repeated observations of the same entities. Fixed effects remove time-invariant entity components and identify coefficients through within-entity variation. Random effects model that component and conventionally require it to be uncorrelated with regressors. The Hausman test compares estimators under specific covariance assumptions; correlated random effects provide another way to examine the independence restriction.
Use for longitudinal firms, banks or organisations when stable unobserved entity differences matter. Choose the estimand and variation source first, then justify fixed, random or correlated random effects substantively.
Strengths
- Fixed effects control additive time-invariant entity differences
- Panel structure separates within-entity change from between-entity differences
Limitations
- Fixed effects cannot estimate effects of time-invariant regressors directly
- Time-varying confounding and dynamic-panel bias can remain
Know the boundary
A Hausman p-value does not decide the scientific design, and a non-significant result does not prove random-effects exogeneity.