Suppose panel data where multiple observations are made of units over time. Regressing a dependent variable measured at each time point on lag of the dependent variable and a unit-specific intercept (a "random effect") - perhaps with an overall intercept for the grand mean - leads to inconsistency and bias of the estimator.
Does this apply in the case of 2 time periods?
Does this apply when the independent variable of interest is randomly assigned? In other words, we are controlling for the lag to improve precision.
