The model is estimated using a random coecients simulated maximum likelihood approach
similar to that summarized in Train (2009).42 The likelihood function for each enrollee is predicted
for a sequence of choices from entry into the Part D program until the end of our data panel.
Details of the methodology are provided in Appendix D.