together with the fact that new enrollees in the program enter the data in every year. These
new enrollees are assumed to choose without inertia; we also assume that the normally distributed
random coecients fully capture the unobserved heterogeneity in their preferences. The parameters
governing the unobserved preference heterogeneity (the distribution of the random coecients) can
therefore be estimated from the choices made by new enrollees in the program. Other determinants
of the decision to switch, most importantly the parameters governing inattention, are identied
from consumers' observed sequences of choices in the years following entry. The initial conditions
problem (e.g., Heckman (1991)23) does not arise in our data because we observe the rst Part D
choices for all individuals in the estimation sample.