International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Switching Costs and Plan Loyalty in Medicare Part C
Adam Atherly, Roger Feldman, Bryan Dowd, Eline van den Broek

Last modified: 28 March 2017



Background: We analyze whether Medicare beneficiaries are sufficiently responsive to changes in health plan premiums and benefits for “premium support” style reforms to be a viable strategy to control Medicare spending. The core belief behind premium support models is that plan participants, if faced with an array of health plan choices, will select the plans that provide the best combination of premium fees and benefit design. This approach, also referred to as “competitive bidding”, is in turn based on two central beliefs: 1) health plans can reduce spending and 2) health plans will have an incentive to pass the savings on consumers in a competitive market. Theoretically, switching costs are a mismatch between a consumer’s current purchase and previous investment. If switching costs are sufficiently large, then the second central belief will be untrue – plans will not have an incentive to pass on saving because plan enrollees will effectively become a captive market.  The policy problem is that if high switching costs effectively lock plan enrollees into their initial health plan choice, then plans have less incentive to produce the optimal benefit package because they effectively have a captive market and in some cases can manipulate consumer preferences.  High switching costs thus can reduce the likelihood that a “premium support” reform could actually produce substantial savings.


This paper estimates the magnitude of switching costs in the U.S. Medicare Part C program. Switching costs in Part C (also termed “Medicare Advantage or MA plans) are interesting for a number of reasons.  First, if prior choice of plan stands in for strong preferences for unobserved plan characteristics, then prior choice is an important omitted variable that may have created bias in previous estimates of own- and cross-price elasticities of health plan choice.  Second, if switching costs are high, this may create incentives for plans to act in non-competitive ways, depending on market structure.  We will also identify individual factors that increase or decrease switching costs, such as age, health and income.

Theory: Consumers are generally assumed to pick plans that provide the combination of benefits and premiums that maximize their individual utility.  However, the plan choice literature has generally omitted prior choices from choice models. In contrast, the influence of prior choices on current choices has received extensive attention in other literatures, such as marketing.  Often referred to as habit persistence or inertia, there is a substantial economics literature on the role of state dependence, i.e., that choices this period are a function of choices in prior periods.  State dependence has been subdivided into two further categories.  Spurious state dependence refers to autocorrelation in the consumer’s utility function error terms over time.  Thus some consumers may serially select certain plans because of unobserved preferences. This then reflects utility maximization – consumers consistently pick a particular plan because they like that plan the best.  In contrast, with structural state dependence, past purchases directly influence the consumer’s choice probabilities for different brands creating “stickiness” in choices that may lead consumers to be in sub-optimal plans.  Sources of structural state dependence include (1) loyalty (a change in preferences due to past purchases, also referred to as psychological switching costs); (2) search costs, which cause consumers not to consider products they have not recently purchased; and (3) learning.


Methods: Our model specifies that the choice in the initial year depends on individual (i) and health plan (j) characteristics in year (t).  Conditional on premiums and benefits, a simplified version of the consumer’s utility index (U) is given by:




where P is the plan’s out-of-pocket premium and B is a vector of benefits.  α and β represent the consumer’s preferences for premiums and benefits.  εijt controls for the consumer’s preference for unobserved characteristics of the jth plan.  Most analyses of health plan choice have used cross-sectional data to estimate models similar to equation (1); analytically, the typical specification uses McFadden’s (1974) conditional logit model.  However, it is not possible to model state dependence with cross-sectional data.  In our data (described below) we observe the same individual making choices over multiple years, and thus we can estimate equation (1) that includes both εijt and a variable representing additional “stickiness” of choice as given by equations (2):





In this model, we have included a “loyalty” factor, L, a dummy variable equal to one if the beneficiary selected plan j in the prior year and zero otherwise.  In each year, the beneficiary is “loyal” to a single health plan – the plan selected the previous year.  However, loyalty in equation (2a) is itself a function of individual characteristics C (equation 2b).  We estimate equations (2a) and (2b) using random coefficients logit, also referred to as “mixed logit” and a latent class model to account for heterogeneous preferences.


Data: The analysis is based on six years (2006-2011) of the Medicare Current Beneficiary Survey, a nationally representative longitudinal dataset.  The MCBS data were combined with data from the Centers for Medicare and Medicaid Services on Medicare Advantage Part C plan benefits and premiums.  Individual choices are modeled as a function of individual characteristics, plan characteristics and prior year plan choices.


Results: We found relatively high rates of switching between plans within insurer (20%), although less switching between insurers.  Prior year plan choices were highly significant at both the contract level (b=4.89, p<.001) and plan level (level (b=1.67, p<.001).  Premium was negative and significant.  Loyalty (contract and plan), premium and plan structure were found to be heterogeneous in preferences, although the results were sensitive to the specification of the distribution of the error term.  Beneficiaries on average need to be compensated $946 to switch plans, controlling for premiums and benefits. The distribution of the loyalty coefficient suggests that 90% of beneficiaries have a positive value for plan loyalty.  We found a statistically significant willingness to pay for a lower prescription drug deductible ($38), for broader networks ($340 for private FFS plans, and $101 for local PPO versus standard MA HMO products) and lower copays, particularly primary care and urgent care (all result significant p<.01).  Loyalty (i.e. plan choice in prior years) was very significant and positive (p<.01).


Conference registration is required in order to view papers.