VoxEU Column Health Economics

The cost of complexity in the Medicare Part D market

Launched in 2006, Medicare Part D allows beneficiaries to enrol in subsidised drug coverage plans sold by private insurers, but navigating the different plans can be complex and lead to sub-optimal choices. This column uses Medicare administrative data for 2006-2010 to understand the quality of consumer decision-making in the Part D marketplace. It finds that the vast majority of elderly place too much weight on premiums relative to out-of-pocket costs, care a great deal about the particular combination of plan features, and are highly likely to choose the same plan every year regardless of changes in prices and alternatives.

The US Medicare programme provides health insurance for the 65-and-over population. The original programme, created in 1965, covered a share of the costs of hospital and doctor visits but failed to cover prescription drugs. A new prescription drug benefit, known as Medicare Part D, took effect in 2006. Under this programme, Medicare beneficiaries could enrol in subsidised drug coverage plans sold by private insurers. The monthly premiums cover about one-quarter of participants drug costs, while Medicare subsidises the rest. 

The large subsidies encouraged the creation of a new insurance market where several private insurers offer many different Part D plans. As Neuman and Cubanski (2009) note, there are about 50 plans to choose from in any given region of the country. And, as Winter et al. (2006) explain, the rules of the programme and of the individual drug plans can be complex. This can lead to substantial cost differences for consumers. For example, Neuman and Cubanski note that “in 2009, patients with Alzheimer’s … taking Aricept could have paid as little as $20 for a month’s supply in one prescription-drug plan or as much as $88 in another”.

Given the complex choice environment created by Part D, an important policy question is whether consumers choose wisely among the many options – in the sense of finding plans that minimise their drug costs – or whether consumers exhibit ‘confusion’ and choose inferior plans. This issue is particularly relevant as many eligible seniors suffer from cognitive limitations due to Alzheimer’s, depression, or other health issues. Furthermore, whether consumers can make good choices in the Medicare Part D marketplace may provide important lessons for the potential efficacy of proposals to allow more consumer choice in healthcare markets generally. It may even shed light on the relevance of the rational choice paradigm in such complex environments.

Do consumers make good choices?

Several studies attempt to evaluate the quality of consumers’ choices in the Medicare Part D market. Winter et al. (2006) calculated that at least three-quarters of people who lacked drug coverage when Part D was introduced could have immediately benefited by signing up for a plan, i.e. the cost savings exceeded the premium. They argued the biggest mistake one could make was to not sign up for Part D at all. In that sense, the roll-out of Part D seemed fairly successful. Both Heiss et al. (2006, 2011) and Levy and Weir (2010) report take-up was high, and the fraction of senior citizens lacking any form of drug coverage fell from about 25% in 2005 to only 7% in 2006. They also report evidence of rational take-up decisions, as seniors who did not sign up for Part D tended to be those with lower drug costs.

Given the high take-up rate, attention has shifted to how well consumers choose amongst the many available Part D drug plans. For instance, Abaluck and Gruber (2011) look at data from 2006 and find that up to 70% of seniors make sub-optimal choices, as they could find a plan generating lower cost without increasing risk. This is perhaps not surprising, as, in 2006, Part D was an unfamiliar new programme. Ketcham et al. (2012) find that those who left the most money on the table in 2006 often switched plans in 2007, leading to substantial savings. But Abaluck and Gruber (2016) argue “there is little improvement in the ability of consumers to choose plans over time”. Thus, the question of how well senior citizens can cope with the complex Part D marketplace remains controversial.

Modelling consumer behaviour

In a further effort to assess the quality of consumer decision-making, Abaluck and Gruber (2011, 2016) estimate models of how consumers choose drug plans. This enables them to learn how consumers value different plan features. They build on two key insights: first, if consumers choose rationally, they should only care about their total costs under a plan, not how this is split between the premium and out-of-pocket costs. Second, consumers shouldn’t care about the combination of plan features – premium, co-pay, deductibles – through which a particular total cost is achieved. Abaluck and Gruber conclude these principles are clearly violated. First, consumers place far too much weight on premiums relative to out-of-pocket costs. Second, consumers do value plans with different co-pays and deductibles differently, even if total costs are the same.     

But Keane and Thorp (2016) argue that the Abaluck-Gruber analysis may be misleading because they assume a ‘one size fits all’ model of consumer behaviour. Presumably, different consumers place different weights on drug-plan features when choosing plans. Suppose most consumers satisfy the two principles of rational-choice behaviour that Abaluck and Gruber suggest, while others do not. An Abaluck-Gruber style of analysis will typically find those principles are violated but miss the point that most individual consumers do follow them. 

Segmenting the market

This concern led my co-authors and I to extend the Abaluck-Gruber analysis to a more general model of choice behaviour that allows for several consumer ‘types’ (Keane et al. 2019). Each type may value plan features differently. Such models are often used in marketing to ‘segment’ consumers so a manufacturer or retailer can design products that appeal to certain types/segments, or target sales promotions at certain types. Here we use market segmentation methods to address a public policy question – i.e. to understand the quality of consumer decision-making in the Part D marketplace. 

We estimate our model of consumer behaviour using Medicare administrative data for 2006-2010.1 We find choice behaviour in the Part D marketplace can be well described by a model with three consumer types. Only 10% of consumers belong to a ‘rational type’ that obeys the two principles proposed by Abaluck and Gruber. About 11% belong to a second type that places far too much weight on premiums relative to out-of-pocket costs. Perhaps they realise that not signing up for Part D is a big mistake, but they don’t really understand the differences between options, so they adopt the heuristic of choosing the cheapest plan they can find. 

The third, and largest, consumer group (79%) belongs to a segment we label ‘confused’. They also place too much weight on premiums relative to out-of-pocket costs. In addition, they care a great deal about the particular combination of plan features through which a given level of cost is achieved. Oddly, they act as if they like higher cost-sharing requirements. This is reminiscent of the finding by Harris and Keane (1999) that many seniors fundamentally misunderstand the cost-sharing requirements of Medicare and behave in the supplemental insurance market as if they like higher cost sharing. The ‘confused’ type also exhibits an extraordinarily high degree of inertia – they are very likely to choose the same plan they chose last year, regardless of changes in prices and costs under that plan or the available alternatives.

How big are consumer losses?

Clearly, only a small segment of consumers can be described as making rational choices. But do the other 90% suffer substantial monetary losses from their sub-optimal behaviour? To address this question, we use our model to simulate how much the ‘confused’ consumers would save if they made choices in the same way as the ‘rational’ type. According to these simulations, they lose an average of $157 per year from sub-optimal behaviour. 

To put this figure in context, we also simulate the consequences of choosing a plan at random. This leads to average losses of $323 per year, which is more than twice as great as the $157 average loss suffered by the ‘confused’ type. Viewed this way, even the ‘confused’ type does much better than ‘throwing darts’. Indeed, one might argue that a mean loss of $157 is quite modest, suggesting the cost of ‘confused’ behaviour is not great. What drives this result is that Medicare subsidises three-quarters of the cost of Part D premiums. Given the large subsidy, even a poorly chosen drug plan is far better than having no plan at all. 

While losses for most consumers from sub-optimal choices in the Part D market appear small, there are notable exceptions. People with Alzheimer’s or depression are more likely to be the ‘confused’ type and they overspend by $394 per year. Our most disturbing finding is that the standard deviation of drug costs for people with Alzheimer’s or depression is $1,240 per year, which is 68% greater than with random choice. This strongly suggests the Part D programme is failing to provide adequate risk protection for some vulnerable groups. 

Can the market be improved?

Finally, we use our model to predict the impact of various policies aimed at simplifying the choice environment. The results of this exercise are disappointing. Streamlining the choice set does little to increase consumer welfare in the Part D market. This is true even if we eliminate the most ‘inferior’ plans, using information that goes beyond what Medicare could plausibly exploit. Thus, although we find clear evidence of sub-optimal behaviour by most consumers, it seems hard to design a government intervention that would meaningfully enhance welfare.


Abaluck, J, and J Gruber (2011), “Choice inconsistencies among the elderly: Evidence from plan choice in the Medicare Part D program”, American Economic Review 101(4): 1180-1210.

Abaluck, J, and J Gruber (2016), “Evolving choice inconsistencies in choice of prescription drug insurance”, American Economic Review 106(8): 2145-2184.

Harris, K, and M Keane (1999), “A model of health plan choice: Inferring preferences and perceptions from a combination of revealed preference and attitudinal data”, Journal of Econometrics 89: 131-157.

Heiss, F, D McFadden and J Winter (2006), “Who failed to enroll in Medicare Part D and why? Early results”, Health Affairs 25: w344-w354.

Heiss, F, D McFadden and J Winter (2011), “The demand for Medicare Part D prescription drug coverage: Evidence from four waves of the retirement perspectives survey”, in D Wise (ed.), Explorations in the Economics of Aging, University of Chicago Press, 159-82.

Keane, M, and S Thorp (2016), “Complex decision making: The roles of cognitive limitations, cognitive decline and ageing”, in J Piggott and A Woodland (eds.), The Handbook of Population Ageing, Vol. 1B, Elsevier North-Holland, 661-709.

Keane, M, J Ketcham, N Kuminoff and T Neal (2019), “Evaluating consumers’ choices of Medicare Part D plans: A study in behavioral welfare economics”, Journal of Econometrics (forthcoming), NBER Working Paper 25652 (March 2019).

Ketcham, J, C Lucarelli, E Miravete and M Roebuck (2012), “Sinking, swimming or learning to swim in Medicare Part D”, American Economic Review 102(6): 2639-2673. 

Ketcham, J, C Lucarelli and C Powers (2015), “Paying attention or paying too much in Medicare Part D”, American Economic Review 105(1): 204-233.

Ketcham, J, N Kuminoff and C Powers (2016), “Choice inconsistencies among the elderly: Evidence from plan choice in the Medicare Part D program: Comment”, American Economic Review 106(12): 3932–3961.

Levy, H, and D Weir (2010), “Take up of Medicare Part D: Results from the health and retirement study”, Journal of Gerontology: Social Sciences 65B(4): 492-501.

Neuman, P, and J Cubanski (2009), “Medicare Part D update: Lessons learned and unfinished business”, New England Journal of Medicine 361: 406-414.

Winter, J, R Balza, F Caro, F Heiss, B Jun, R Matzkin and D McFadden (2006), “Medicare prescription drug coverage: Consumer information and preferences”, Proceedings of the National Academy of Sciences 103(20): 7929-34.


[1] The administrative data were assembled in Ketcham et al. (2016). They include detailed information on drug purchases, health conditions, and Part D plan choices. The Part D plan cost calculator developed was developed in Ketcham et al. (2015). It uses person-specific prescription information to yield accurate estimates of the mean and variance of out-of-pocket costs of each person

525 Reads