Panel Data

Panel data (longitudinal data) tracks the same units — people, firms, states, judges — across multiple time periods. Each unit appears in the dataset more than once. This unlocks comparisons of a unit to itself over time, which controls for every stable, time-invariant characteristic of that unit (observable or not). The structure is indexed by $(i, t)$ with $i$ for unit and $t$ for time.

When to use

Reach for panel methods whenever the data have repeat observations per unit. The standard moves: Fixed Effects with unit dummies to absorb time-invariant unobservables, Time Fixed Effects with time dummies for nationwide shocks, Two-Way Fixed Effects for both, and Clustered Standard Errors (clustered at the unit level) for honest inference under within-unit serial correlation. PS_4 is the canonical example: 48 states × 23 years of seatbelt-law data, where the staggered timing of state-by-state law adoption supplies the treatment variation.