POL222 - week 11

When can regression be interpreted casually?

Must be all of the following; No reverse causality, all relevant confounders controlled (problem: some are unobservable), correct functional form (model correctly estimates conditional means)

Types of Observational Data

Cross-sectional: many units, one time period.

Time series: one unit, many time periods.

Panel (TSCS): many units over many time periods, allows between-unit and within-unit comparisons

Unit Fixed Effects Model

Focuses on within-unit over time variation, controls for all stable unit characteristics, observed or unobserved, coefficient of X = how Y changes within units as X changes, net of all fixed traits

Why Multiple Regression is essential

Allows control for many confounders simultaneously, allows Z to take many possible values, produces unbiased estimates if all relevant confounders included

Still limited by unobservable confounding and model assumptions