Modeling and Evaluating Consumers' Prescription Drug Plans Choices in Medicare Part D

During the first five years of Medicare Part D the average beneficiary could have saved between $292 and $520 by purchasing the same drugs under a different prescription drug plan (PDP). What caused this persistent overspending? Are people too confused about how Part D works to choose their cost-minimizing plans? Or are they simply choosing to pay for higher quality plans that offer better customer service and risk protection? The answers to these questions are critical for assessing the cost and benefits of policy reforms that have been proposed to simplify the Part D program. Past proposals have included standardizing certain aspects of plans, limiting the number of insurers, limiting the number of plans each insurer can offer, and providing individualized "nudges" to people.

In this project we capitalize on the ideal CMS data and a well-validated cost calculator to observe what each person could have spent in each plan available to her from 2006-2010. These data also provide us with a rich set of plan-specific attributes as well as time-varying measures of health from Medicare Parts A and B claims data. With these data we observe that overall plan retention rates are fairly low: by 2010, only half of those who had previously enrolled in a PDP were still in their original PDP. Further, switching between PDPs was responsive to the individuals' relative costs of their status quo plans. Contrary to choice overload, facing larger choice sets did not inhibit the likelihood that people switch PDPs. 

We also move beyond these descriptive results to evaluate choice quality directly. Using well-established non-parametric approaches, we find that 25% of PDP enrollees chose a dominated plan in 2006, but this declined to 10% by 2010. Finally, we explore how well standard logit models perform in explaining and predicting individuals' PDP choices. To accomplish this we conduct a series of internal and external validation tests. Poor performance of such models will point out the need for researchers to develop alternative approaches to understanding and evaluating consumers' choices of prescription drug plans.