The Oxford Research Encyclopedia of Economics and Finance is moving into subscription mode on April 30. Visit About to learn more, meet the editorial board, or learn how to subscribe.

Show Summary Details

Page of

PRINTED FROM the OXFORD RESEARCH ENCYCLOPEDIA, ECONOMICS AND FINANCE ( (c) Oxford University Press USA, 2020. All Rights Reserved. Personal use only; commercial use is strictly prohibited (for details see Privacy Policy and Legal Notice).

date: 31 March 2020

Choice Inconsistencies in the Demand for Private Health Insurance

Summary and Keywords

In many countries of the world, consumers choose their health insurance coverage from a large menu of often complex options supplied by private insurance companies. Economic benefits of the wide choice of health insurance options depend on the extent to which the consumers are active, well informed, and sophisticated decision makers capable of choosing plans that are well-suited to their individual circumstances.

There are many possible ways how consumers’ actual decision making in the health insurance domain can depart from the standard model of health insurance demand of a rational risk-averse consumer. For example, consumers can have inaccurate subjective beliefs about characteristics of alternative plans in their choice set or about the distribution of health expenditure risk because of cognitive or informational constraints; or they can prefer to rely on heuristics when the plan choice problem features a large number of options with complex cost-sharing design.

The second decade of the 21st century has seen a burgeoning number of studies assessing the quality of consumer choices of health insurance, both in the lab and in the field, and financial and welfare consequences of poor choices in this context. These studies demonstrate that consumers often find it difficult to make efficient choices of private health insurance due to reasons such as inertia, misinformation, and the lack of basic insurance literacy. These findings challenge the conventional rationality assumptions of the standard economic model of insurance choice and call for policies that can enhance the quality of consumer choices in the health insurance domain.

Keywords: health economics, private health insurance, risk aversion, behavioral mistakes, cognitive limitations, choice overload, inertia, adverse selection, economic policies, welfare

Choice Complexity in Private Health Insurance

In many countries of the world, consumers choose their health insurance coverage from a large menu of options supplied by private insurance companies. The United States, where private insurance is the dominant form of health insurance for the majority of individuals, is perhaps the most prominent example. Increasingly the beneficiaries of Medicare and Medicaid, two major public health insurance programs in the United States, also choose their healthcare plans from a large number of options provided by private insurers under contract with the government (Gruber, 2017). For example, Medicare beneficiaries considering enrollment into a Medicare Part D prescription drug plan face a choice set of at least 30 plans with different premiums and cost-sharing characteristics offered by several private insurers in their area (Polyakova, 2016). In a typical U.S. county, consumers searching for a health insurance plan through the health insurance exchanges established by the Patient Protection and Affordable Care Act of 2010 choose from 46 health plans with different levels of coverage offered by five different insurers (Ericson & Sydnor, 2017).

Switzerland implemented a system of managed competition on the health insurance market in the 1990s. Under this system, basic health insurance coverage is mandated by law and basic health plans are provided by subsidized private insurers who compete in the market subject to strict regulations of contract design, pricing, and medical underwriting. Despite these homogenizing regulations, consumers in a typical Swiss canton choose their standard plan from more than 30 options that vary in price and other characteristics (Frank & Lamiraud, 2009). Since the 1990s, Belgium, Israel, Germany, and the Netherlands have similar systems of managed competition on the market for health insurance where private insurers compete for customers within the constraints of regulation (Schmueli, 2015), and consumers face a large menu of health insurance coverage options with different prices and contract designs.

Australia has a publicly funded universal healthcare system that provides free of charge medical care to public patients in public hospitals. To bypass public hospital waiting lists, patients can purchase supplementary private health insurance that covers treatments of private patients in public and private hospitals. Prices of these supplementary insurance plans are subject to community rating and medical underwriting is not allowed; therefore, insurers have incentives to compete on product design. An Australian consumer choosing a supplemental health insurance may face a choice set with 48 to 2,050 options that differ along a range of attributes such as premiums, excess, covered treatments, and co-insurance (Kettlewell, in press).

There are potential benefits from giving consumers a wide choice of health insurance options. With more options, consumers are more likely to find insurance plans that meet their preferences and healthcare needs, while insurers competing for customers will have incentives to create higher quality and more affordable products. However, these benefits rest on the assumption that consumers are active, well informed, and sophisticated choosers who can think through costs and benefits of a large number of health insurance options with complex design and choose plans that are well-suited to their individual circumstances, such as expected future health, willingness to tolerate risks, and preferences for medical providers.

During the last two decades of the 20th century economists and health services researchers have been accumulating evidence that consumers’ ability to make rational decisions in the context of health insurance may be limited. For example, Harris and Keane (1998) have shown that elderly Medicare beneficiaries have inaccurate beliefs about cost-sharing characteristics of health insurance options in their choice set that included basic Medicare without supplemental coverage, supplemental Medigap plans with and without prescription drug coverage, and two different HMO plans. These inaccurate perceptions had significant effects on consumers’ health insurance choices. This study corroborates the earlier findings from health services research (HRS) about a lack of understanding of key features of basic Medicare and supplementary health insurance plans by elderly Americans (Cafferata, 1984; Davidson, Sofaer, & Gertler, 1992; McCall, Rice, & Sangl, 1986). Fang, Keane, and Silverman (2008) analyze the determinants of holding a supplementary Medigap health insurance policy by Medicare beneficiaries in the HRS survey. They show that Medigap enrollees are advantageously selected (i.e., have lower medical expenditure risk than those without supplemental coverage) and that this advantageous selection can be attributed to a negative correlation between cognitive ability and health expenditure risk. These findings were subsequently corroborated in Keane and Stavrunova (2016) who analyzed the demand for Medigap within a comprehensive econometric framework, disentangling selection from moral hazard in this market. One interpretation of these results is that unhealthy consumers who would presumably benefit more from supplementary health insurance coverage cannot make effective decisions in the highly complex Medigap market due to a lack of cognitive capacity necessary to make such decisions.

Many prominent economists pointed out the emerging evidence of consumer confusion as grounds for serious concerns about the desirability of outcomes arising from unrestricted consumer choices in the health insurance markets (e.g., McFadden, 2006; Frank & Zeckhauser, 2009; Liebman & Zeckhauser, 2008). In the first two decades of the 21st century, an increased availability of large-scale detailed administrative data with information about characteristics of health insurance options in consumers’ choice sets, actual insurance choices, and subsequent claims, in some cases linked to survey data on relevant perceptions and beliefs of consumers, allowed researchers to quantify financial and welfare losses from inefficient choices of health insurance.

This article surveys recent studies quantifying financial and welfare consequences of consumer confusion regarding the market for private health insurance. It builds on earlier comprehensive surveys of literature on health insurance markets such as Cutler and Zeckhauser (2000), Cutler and Zeckhauser (2004), Einav, Finkelstein, and Levin (2010), McGuire (2012), and Keane and Thorp (2016). The article is organized in the following sections: “Demand for Health Insurance: Theory” lays out a simple model of health insurance plan choice with and without demand frictions and discusses implications of demand frictions for welfare analysis; “Choice Inconsistency in Medicare Part D Prescription Drug Insurance” provides a selective review of studies on choice inconsistencies in the market for Medicare Part D prescription drug insurance; “Choice Inconsistencies in Employer-Provided Insurance” reviews studies on demand frictions in the setting of employer-provided health insurance; “Concluding Remarks” completes the article.

Demand for Health Insurance: Theory

In the standard economic model of the demand for insurance, a risk-averse individual values insurance insofar as it reduces the riskiness of future wealth. In the modern health insurance context, health insurance plans not only provide financial protection from medical expenditure risks, but can also determine the type of medical care the consumer can access. In particular, plans can cover services of different healthcare providers, access to different sets of treatments and prescription drugs, etc. Health insurance plans can also differ in the degree of administration burden they impose on enrollees. These non-financial plan attributes may impact consumers’ utility above and beyond the risk-reducing properties of the plans’ financial attributes. With these considerations in mind, this section formulates a choice model for a rational risk-averse individual who chooses a health insurance plan from a set of options, J, to maximize the expected utility which depends on the plan’s financial and non-financial attributes. In particular, the expected utility of individual i in plan j, uij, is given as follows:


where wi denotes an individual’s financial wealth, pj is plan j’s premium, e denotes the out-of-pocket medical expenditure, λj are non-financial characteristics of the plan that directly affect utility above and beyond the plan’s impact on out-of-pocket medical expenditure, and θj are the plan’s financial attributes (such as premiums, deductible, and co-insurance rates) that change the distribution of the individual’s out-of-pocket medical expenditure, e, but do not affect utility beyond their effects on out-of-pocket cost. The function fi(e|θj) is the probability density function of out-of-pocket medical expenditure for individual i insured by health plan j. The financial attributes, θj, affect the distribution of out-of-pocket costs of individual i by changing its mean, variance, and possibly other moments. Given plan j, the distribution of out-of-pocket costs can vary between individuals due to differences in healthcare needs and other characteristics. Finally, the parameter ξi denotes risk aversion of individual i.

The individual faces uncertainty about her future healthcare needs. She does not know for sure which health state will be realized in the future, but she knows the distribution of possible health outcomes. This knowledge, combined with full information about prices of medical care and understanding of cost-sharing properties of health plans in her choice set, allows the consumer to construct the distribution of out-of-pocket expenditures for each plan j. The individual will choose plan j that gives her the highest expected utility, uij. Ceteris paribus, an individual would prefer plans that result in lower mean and variance of out-of-pocket expenditures and plans with more valuable attributes, λj.

Under specific functional form assumptions, this model gives rise to an empirical specification that can be used to estimate the primitive parameters of the utility function from administrative data on plan choices and insurance claims of individuals. Notable examples of such analysis in the context of employer-provided health insurance are Handel (2013) and Handel and Kolstad (2015), while Cohen and Einav (2007) develop a similar framework to study consumers’ choices of auto insurance.

This model can also be used to motivate a “reduced form” empirical discrete choice model of consumers’ plan choice where an indirect utility of a plan can be specified as a flexible function of a plan’s attributes without making functional form assumptions about the primitives of the model. Consumers’ hedonic valuation of a plan’s attributes is then estimated using logit specification. Notable examples of this approach include Harris and Keane (1998), Bundorf, Levin, and Mahoney (2012), Abaluck and Gruber (2011), Polyakova (2016), and Ericson and Starc (2016).

Both approaches can be used for welfare analysis, such as assessing welfare costs of adverse selection and welfare impacts of counterfactual policies. In addition, the first approach can yield estimates of the population distribution of risk aversion which is of independent interest for many areas of economics research. Einav et al. (2010) provide a comprehensive review of both approaches.

In the second decade of the 21st century, several influential studies demonstrated that actual choices of consumers in health insurance markets are often not consistent with the assumptions of rational utility maximizing behavior described by Equation 1. In particular, there is growing evidence that consumers “leave money on the table” by making inefficient and even dominated choices of health plans due to inertia, misinformation about plan attributes, lack of understanding of basic insurance concepts, and other cognitive and behavioral demand frictions. These studies suggest that a more realistic model of health insurance choice would allow the consumer’s decision process to depart from the rational framework (Equation 1) along some relevant dimensions. One way to model these demand frictions is to explicitly specify that consumers base their choices on the parameters λj,^, θj,^ and fi(.)^ that may differ from their correct (objective) counterparts λj,, θj, and fi(.) due to inaccurate subjective beliefs, lack of competence, and other cognitive and behavioral reasons:1


If the true decision process for insurance choice is given by Equation 2, the empirical model based on Equation 1 will not recover the correct distribution of preferences for risk and plan attributes from the usual administrative data on plan choices and claims without information about consumers’ subjective beliefs and competences. For example, Handel and Kolstad (2015) show that such a model overestimated risk aversion in the setting where consumers’ choices of employer-provided health insurance were prone to inertia and affected by biased subjective beliefs about the plans’ non-financial attributes.

Furthermore, welfare analysis of insurance markets with demand frictions is challenging. Biased beliefs about plans’ attributes, lack of insurance competence, and other limitations may determine plan choices according to Equation 2, but the ex-post utility from the plan is very plausibly determined by the objective plans’ characteristics and is given by Equation 1. For example, if a consumer incorrectly believes that one prescription drug plan has a richer drug formulary than another plan, this belief will affect her plan choice but not the actual utility experienced in the two plans. Spinnewijn (2017), Handel and Schwartzstein (2018), and Handel et al. (2018) demonstrate that the demand frictions create a wedge between the true value of insurance to consumers and their willingness to pay for it revealed by their choices. In this setting, the insurance demand estimated from a consumer’s revealed choices is no longer relevant for welfare analysis as it does not reflect the consumer’s true valuation of insurance.

For a meaningful welfare analysis, the revealed demand curve should be adjusted for the wedge that the demand frictions drive between the willingness to pay for insurance and its value to consumers. A variety of approaches have been developed in the literature to quantify this wedge and to construct the welfare-relevant valuation curve. For example, researchers may estimate the welfare-relevant valuation curve from data on choices of experts as it is reasonable to assume that the decisions of experts are not affected by demand frictions (e.g., are consistent with Equation 1). Alternatively, researchers may use surveys to identify informed and uninformed consumers and use choices of informed consumers only to estimate the valuation curve (e.g., Handel & Kolstad [2015] and Handel et al. [2018]). Finally, researchers can remove the frictions by conducting informational intervention and estimate the welfare-relevant valuation curve using data on choices made after the intervention (e.g., Bhargava, Loewenstein, & Sydnor, 2017). Handel and Schwartzstein (2018) provide a thorough discussion of assumptions underlying these approaches.

The sections “Choice Inconsistency in Medicare Part D Prescription Drug Insurance” and “Choice Inconsistencies in Employer-Provided Insurance” of this article survey selected studies that document inconsistencies in the actual consumer choices in the contexts of Medicare Part D prescription drug insurance and employer-provided health insurance, quantify financial and welfare losses arising from these inconsistencies, and identify several reasons behind poor consumer choices in the health insurance domain.

Choice Inconsistency in Medicare Part D Prescription Drug Insurance

Medicare Part D—Institutional Framework

Medicare in the United States is a national health insurance program for senior individuals that began in 1966. Prior to 2006, Medicare provided coverage for hospital and physician service but did not cover prescription drugs. In 2006, as a part of the Medicare Modernization Act of 2003, Medicare was expanded to include prescription drug coverage, and the Medicare Part D prescription coverage program was launched.

The important innovation of the Medicare Part D program is that it is delivered by private insurers under contract with the government. Private insurers can offer various Part D plans provided that these plans have actuarial value equal to or greater than that of the Standard Defined Benefit (SDB) plan. In 2006, the SDB plan had a relatively low deductible of $250, a constant co-insurance rate of 25% up to the initial coverage limit (ICL), a 100% co-insurance rate up to the Catastrophic Coverage Limit (CCL) of $5,100 (the “donut hole”), and a 5% co-insurance rate on expenditures above the CCL. In later years, the ICL and CCL have somewhat decreased, making the SDB more generous, but the basic structure of the SDB remained unchanged.

Private insurers have been actively participating in the Medicare Plan D market from the start of the program. By the end of 2006, there were more than 3,000 Part D plans available to potential enrollees in the entire country, and a typical U.S. county had 48 Part D plans for the enrollees to choose from (Abaluck & Gruber, 2011). Some companies offer only one plan in a given geographic market, but it is not unusual for insurers to offer several plans of different generosity and premium levels in the same market.

Polyakova (2016) demonstrated that in the period 2006–2009, Part D plans could be roughly classified into four types: Type 1 offered the SDB level of deductible and no coverage of the donut hole; type 2 offered a lower deductible than the SDB but no donut hole coverage; type 3 offered a lower deductible than the SDB and partial coverage of the donut hole; and the most generous type 4 offered a lower deductible and full coverage of the donut hole. The plans also differed along other attributes (e.g., provider networks, drug formularies) and premiums. In 2007, the average annual premiums ranged from $216 for a type 1 contract to $610 for a type 3 contract. The most generous and increasingly most expensive type 4 plans (average price $668 in 2006 and $1,291 in 2007) were not offered in the market, starting from 2008 (Polyakova, 2016).

Medicare Part D enrollment is voluntary and the market is highly regulated. Potential enrollees cannot be denied coverage by any plan, and all beneficiaries (incumbent and new) in a given plan pay the same price (i.e., there is no medical underwriting) and can switch to a different plan during the open enrollment period. Enrollees in plans that were changed or consolidated are automatically re-enrolled by the Centers for Medicare and Medicaid Services (CMS) in default plans in the absence of active re-enrollment decisions. There are several risk-adjustment policies in place to combat adverse selection in the Part D market.

Choosing a Medicare Plan D plan is a challenging task for many potential enrollees. Consumers should compare 30 or more plans on many dimensions and map the complex cost-sharing characteristics into out-of-pocket expenditures, taking into account uncertainty about their own future health status. To make optimal decisions, consumers must understand the implications of cost-sharing aspects of the plans for out-of-pocket expenditures, be able to correctly forecast their future prescription drug needs and costs, and be well-informed about non-financial aspects of plans that are important to them. If any of these requirements is not satisfied, the outcome of the choice process is likely to be suboptimal.

Choice Inconsistencies in Medicare Part D Prescription Drug Choices

The work of Daniel McFadden and colleagues examining intentions and preferences of consumers in the wake of the launch of Medicare Plan D gave early warning signs that consumers may have inadequate expertise to benefit from the vast choice available in this market (Heiss, McFadden, & Winter, 2010; McFadden, 2006; Winter et al., 2006). The researchers carried out an online panel survey (Retirement Perspective Survey [RPS]) that collected rich information on the characteristics, preferences, and intentions of elderly Americans, with a specific focus on their experiences with Medicare Part D in the years 2005, 2006, and 2007. One module of the survey presented the respondents with a hypothetical Part D plan choice problem. Only a minority of consumers (36%) chose cost-minimizing plans in this setting, and the choices revealed that consumers do not value risk-reducing attributes of more comprehensive plans. In one of the contributions, McFadden concluded: “The new Medicare Part D prescription drug insurance market illustrates that leaving a large block of uninformed consumers to sink or swim, and relying on their self-interest to achieve satisfactory outcomes can be unrealistic” (McFadden, 2006, p. 23).

It is possible, however, that individuals would behave differently and choose more efficiently in a real-choice environment where the financial consequences of poor choices are more damaging. Abaluck and Gruber (2011) analyzed the quality of actual choices in the Medicare Part D market in 2006, the year of the program launch. In particular, they tested whether consumers’ choices in this market were consistent with the rational utility maximizing choice behavior under full information. In the standard model of the demand for health insurance, rational risk-averse consumers value plans primarily for their ability to reduce mean and variance of future healthcare expenditures. These costs consist of premiums known with certainty at the time of the choice, and the ex-ante uncertain out-of-pocket costs determined by financial characteristics of the plan and future realization of a consumer’s health status. The results of Abaluck and Gruber’s (2011) study suggested that seniors’ choices of Medicare Part D plans are inconsistent with the rational choice behavior along several dimensions. In particular, (a) in choosing health insurance, consumers value plan premiums and out-of-pocket costs differently, placing higher weight on premiums; (b) consumers value a plan’s financial attributes beyond their impact on out-of-pocket costs; and (c) consumers place little value on characteristics of plans that reduce the variance of out-of-pocket medical costs.

The study used a unique data set with detailed information on prescription drug use for about 0.5 million Medicare beneficiaries, linked to information on all Medicare Part D plans and their attributes that were available in each consumer’s area (i.e., consumer’s choice set) from CMS public files. To construct the prospective distribution of prescription drug costs for each plan in the consumer’s choice sets, the authors matched each consumer in the data set to 200 “identical” consumers, determined by quantiles of several drug expenditure categories in 2005, the year preceding the choice of Medicare Part D plan. The realized drug claims of these matched individuals in 2006 (after Part D plans were chosen), combined with the plans’ cost-sharing characteristics, were used to approximate the distribution of out-of-pocket costs for all plans in consumers choice sets. The authors then tested whether consumers’ plan choices were consistent with rational and informed choice under uncertainty characterized by these distributions.

The descriptive analysis suggested that only 12.2% of consumers chose a Part D plan that minimized their total out-of-pocket costs, with consumers losing at least 30% of their Part D spending by choosing a suboptimal plan. This choice inefficiency cannot be rationalized by risk aversion because, for the majority of consumers, the lower-cost plans had the same or lower variance than their chosen plan.

The authors showed that the main reason for these choice inefficiencies was that consumers valued financial characteristics of the plans beyond their impact on out-of-pocket expenditures. To this end, they specified a conditional logit model of Part D choice where individual i’s utility from choosing plan j was given by


where πj denotes plan j’s premium, μij* and σij2 are the mean and variance of out-of-pocket cost for individual i on plan j, xj is a vector of financial plan characteristics (i.e., deductible, generosity of donut hole coverage, and the cost-sharing index), qb(j) is a vector of non-financial plan characteristics that vary across the brands (i.e., generosity of drug formularies and quality index), and ϵij is identically and independently distributed (i.i.d.) according to the extreme value type 1 distribution. To alleviate concerns about unobserved plan attributes, the model also included a rich set of variables, capturing all publicly available information about plans, and a full set of interacted brand-state fixed effects.

The authors showed that under the assumption of a rational informed utility-maximizing plan choice, the parameters of the logit model should satisfy the following three restrictions: (a) Controlling for the risk characteristics of the plan, the coefficient on premiums should be equal to the coefficient on expected out-of-pocket costs 0=β1) (i.e., “the individuals should be willing to pay exactly one dollar in additional premiums for coverage, which reduces expected out-of-pocket costs by one dollar” [Abaluck & Gruber, 2011, p. 1194]). If this restriction does not hold, the individuals are not choosing efficiently (i.e., they could switch to a plan with comparable risk characteristics but with lower expected costs); (b) conditional on a plan’s premiums and mean and variance of out-of-pocket costs, the additional financial characteristics of the plan should not matter for choice =0); (c) the coefficient on the plan-specific variance of out-of-pocket costs is negative 2<0) (i.e., the individuals are risk-averse and value variance-reducing characteristics of plans). It turns out that all three restrictions are violated in the empirical model estimated using data from the actual plan choices of Part D enrollees. Consumers appeared to value lower premiums much more than they valued an equivalent reduction in out-of-pocket costs. They valued financial plan characteristics above and beyond the effects of these characteristics on the distribution of out-of-pocket costs. In particular, they were willing to pay for a lower deductible, more generous cost sharing, and donut hole coverage beyond the impacts of these attributes on their out-of-pocket costs. Finally, consumers undervalued the risk-reducing aspects of the plan—the coefficient on the variance of out-of-pocket costs was negative but also very small and not statistically different from zero in some specifications. The first two conclusions are robust to controlling for private information and measurement error, while the last conclusion is more fragile and depends on how the expenditure variance is measured.

The analysis does not explore the reasons behind inefficiency in consumer choices of Part D plans. The authors conjectured that, given the complexity of the choice problem in this market, consumers may have been intentionally relying on some trusted heuristics in their choices instead of carrying out the full cost-benefit analysis of all options. Alternatively, the consumers chose inefficiently due to their lack of basic understanding of health insurance.

Regardless of their reasons, the welfare analysis suggests that if the choice inconsistencies were removed (i.e., if the three restrictions stemming from the rationality assumption were imposed on the parameters of the empirical choice model), consumer welfare would increase by 27% in a partial equilibrium, without the endogenous plans’ repricing. The authors suggested that there may be a large scope for government intervention in Plan D in the form of targeted information provision. Additionally, decreasing the number of plans in consumers’ choice sets to simplify the choice process can also be welfare improving under reasonable assumptions.

Heiss et al. (2013) assessed quality of consumer choices and estimated savings that would result if suboptimal choices were eliminated, using a representative, large-scale administrative data set on 20% of Medicare beneficiaries that included detailed information on Medicare claims, including Part D, for 2006–2008. As in Abaluck and Gruber (2011), the individual-level data was supplemented by detailed information about characteristics of Part D plans in each individual’s choice set, provided by the CMS. Using sophisticated algorithm, the authors simulated consumers’ prospective out-of-pocket costs associated with every plan in their choice set using detailed information on their past prescription drug claims and premiums and risk-sharing characteristics of plans. The researchers considered several reasonable benchmark decision rules to which actual consumer choices can be compared (e.g., choices based on the lowest premium, on the advice of the Medicare Part D online decision-support tool Prescription Drug Plan Finder, and under perfect foresight about prescription drug needs). The analysis revealed that each of these benchmarks would have resulted in more cost-effective plan choices for a majority of consumers. The consumers could have saved up to $197 per year if they chose plans based on the advice of the Plan Finder and up to $115 if they chose plans with the lowest premium. The study concluded that “At least for Medicare Part D plans with their rather complex structure, our results do not support the proposition that consumers can make and benefit from good choices in private health insurance markets, and ‘vote with their feet’ to direct healthcare resources to their best use” (Heiss, Leive, McFadden, & Winter, 2013).

It is possible that choice inconsistencies documented in these contributions were due to consumers’ initial lack of familiarity with the new program. As consumers learn about the program through hands-on experience, they may over time switch to more suitable plans. Several articles investigated whether consumer choices of Medicate Part D plans improved in the years following initial enrollment. Ketcham, Lucarelli, Miravete, and Roebuck (2012) investigated the dynamics of choices of Part D plans. They matched longitudinal administrative data from a pharmacy benefit manager, CVS, on prescription drug claims of close to 70,000 individuals continuously enrolled in a Part D plan in 2006 and 2007 with the CMS information about characteristics of plans in consumers’ choice sets. They assumed that consumers were fully informed about their next year’s drug needs, estimated costs under alternative health plans using ex-post drug consumption, and showed how “overspending” (the difference in out-of-pocket spending between the chosen plan and the person’s cheapest alternative) changed in the first year following introduction of Medicare Part D.

Their analysis suggests that mean overspending fell by $300 in 2007 from $546 in 2006 (54% reduction) in their data, with 81% of individuals in the sample decreasing their overspending, and 19% increasing it slightly. Females, older individuals, individuals who started treatment for Alzheimer’s, and those overspending the most in 2006 decreased their overspending more than other groups, suggesting that, at least to some degree, this improvement was facilitated by family members, social networks, medical personnel, and other institutions. The primary source of improvement in overspending was switching plans, with switching being more likely after a significant overspending. Around 54% of individuals in the sample switched their plan between 2006 and 2007. These results contrast with the evidence of consumer inertia in other insurance contexts and are consistent with consumers learning about the new market.

Abaluck and Gruber (2016a) addressed the same question as Ketcham et al. (2012) but with a superior data set that contained information on prescription drug utilization and claims for 20% of the entire population of Plan D enrollees in the period 2006–2009, matched with detailed information about attributes of Part D plans in each consumers’ choice set from the CMS files. Similar to their earlier study, the authors estimated the distributions of prospective costs for each plan in consumers’ choice sets assuming (a) perfect foresight (i.e., the individuals knew their next year’s drug consumption when they were choosing the plan), and (b) “rational expectations” (i.e., individuals’ expected spending next year was computed based on their drug consumption in the previous year).

Like previous studies, Abaluck and Gruber (2016a) showed that a large majority of enrollees (85%) did not choose the lowest cost plan when they first enrolled in Part D. However, in contrast to Ketcham et al. (2012), in these data a much lower proportion of consumers (less than 10%) switched plans each year, suggesting substantial inertia in plan choices, and the proportion of individuals in a cost-minimizing plan fell over time, with only 2% of individuals having the lowest cost plan by 2009 (3 years after introduction of the program). Using a similar approach to welfare analysis as in Abaluck and Gruber (2011), the authors showed that consumers’ welfare loss from suboptimal choices increased over time, and that supply-side factors such as changes to premiums, plan generosity, and plan exit were major contributors to this change.

Several studies investigate why consumers make poor choices in the Medicare Part D market. Kling, Mullainathan, Shafir, Vermeulen, and Wrobel (2012) is an experimental study that pointed out informational frictions as one possible reason for choice inefficiencies in this market. The authors conducted a field experiment in which consumers in a treatment group received a letter with information about the costs of Part D plans personalized for their drug profiles. In particular, the letter stated (a) the predicted prescription drug costs of their current Part D plan, conditional on their drug profile; (b) the predicted drug costs in the lowest cost plan, conditional on their drug profile; and (c) predicted savings from switching to the lowest cost plan. This information was also available to consumers free of charge on Medicare’s website decision support tool, the Prescription Drug Plan Finder. The control group was given the address of the Plan Finder website but not information about the costs. If consumers efficiently utilized all available information when making the choice of a Part D plan, the letters should not have affected consumers’ plan choices. Surprisingly, however, providing consumers with the relevant information in the mail rather than having them access such information themselves from publicly available sources made a difference in their choices. Among the treatment group, 28% of individuals changed their plan in the following year compared to 17% in the control group. Plan switching in the treatment group resulted in more efficient choices, leading to a 5% reduction in out-of-pocket costs relative to the control group. The less-educated individuals benefited more from the informational intervention, which is consistent with the earlier studies on the importance of cognitive limitations for health insurance choice.

Ho, Hogan, and Morton (2017) hypothesized that consumers’ inertia and resultant inefficient choices arose because consumers did not pay attention to their Part D choice set all the time but only when prompted by changes in circumstances. They developed a structural model to estimate the effect of consumer inattention on switching behavior in the Part D plans market and on plans’ prices. Researchers used administrative data from the CMS on Medicare Plan D plan choices of about 250,000 individuals over the 2006–2009 period. The data are extremely rich in that they contain information on demographic and health status characteristics of beneficiaries, characteristics of all Part D plans available in all 34 Medicare regions of the United States, Part D plans chosen by the beneficiaries, and the ex-post costs of their prescription drug purchases. They show that Part D plans’ premiums were on average 62% higher in 2009 than they were in the year of the program’s launch in 2006, despite very modest increases in insurers’ costs of providing these plans. Similar to previous studies on Medicare Plan D, the authors demonstrated that consumers left significant amounts of money on the table by not switching to the lowest cost plans. They also showed that consumers are significantly more likely to switch plans after experiencing a shock to their plan’s premiums and coverage or after an acute health event. These observations motivated a two-stage modeling approach which allowed separating the effect of consumer inattention from other determinants of inertia. In the model, consumers did not actively search for a new plan (were inattentive) until they experienced a sufficiently large shock to their awareness about alternatives to their current plan. In particular, in their model the shock to consumer i’s awareness at time t vi,t is given by


where vi,k,t (k = p,c,h) are the shocks to the consumer’s current plan’s premiums, coverage, and health, while vi,e,t captures other determinants of awareness not observed by an econometrician (e.g., advice from a younger relative). The coefficients βk measure relative importance of the shocks on the consumer’s awareness. The consumer became aware and decided to re-optimize her Part D plan choice if the composite shock, vi,t, exceeded the threshold vi.t~, which varied with consumers’ demographics and time period.

Once the consumer decided to re-optimize the choice of a Part D plan, she would consider all plans in her choice set, including the current plan, and would evaluate these plans according to the following utility function:

ui,j,t=OOPi,j,t^α1+Premiumj,t[ α2,1+vi,p,tα2,2 ]+Deductiblej,tα3,1+Gapj,t[ α4,1+vi,c,tα4,2+vi,h,tα4,3 ]+Xj,tα5,i+ϵi,j,t,

where OOPi,j,t are the expected out-of-pocket costs estimated using the approach similar to that in Abaluck and Gruber (2011) and Ketcham et al. (2012), and the effects of a plan’s premium and donut hole coverage (Gap) depend on the magnitude of a shock to a consumer’s awareness due to changes in the plans’ characteristics and health. This specification implies that removing inattention will increase the elasticity of demand with respect to premiums. The vector Xj,t includes non-financial plan characteristics and brand fixed effects, and the persistent unobserved heterogeneity in plan preferences is modeled by specifying random coefficients on brand fixed effects. The random term ϵi,j,t follows the i.i.d type 1 extreme value distribution. The estimation results suggest that in the years following the program launch, only 37% of consumers are attentive (i.e., make active plan choices).

Ho et al. (2017) then showed that in the 2006–2009 period, the market for the Part D plans in New Jersey had been characterized by oligopolistic competition, significant price dispersion for similar plans, and increasing concentration, product differentiation, and plan premiums. The distinctive feature of the analysis is that the pricing decisions of insurers are explicitly affected by the degree of consumer inattention. Higher inattention lowers consumers’ sensitivity to insurance premiums and makes the demand less elastic, allowing insurers to raise premiums without losing market share. The counterfactual simulations estimate the effect of removing the inattention friction on plan premiums and consumers’ out-of-pocket costs. Making all consumers attentive leads to total savings of $1,154.20 per enrollee over 3 years (25.6% of total spending) via decreases in out-of-pocket costs and premiums. Premiums decline because the insurers facing more elastic demand decrease prices strategically and because price-sensitive consumers prefer less expensive plans. Ho et al. (2017) abstract from adverse selection issues by assuming that the risk adjustment subsidies to insurers implemented in the Part D market are effective.

Polyakova (2016) is the first study that investigated the effects of consumer inertia in Part D plan choices on adverse selection in this market. She used rich administrative data from the CMS on Medicare Plan D plan choices of about 1 million individuals over the 2006–2009 period. Certain rigidities in plan offerings across regions and over time allowed classification of all prescription drug insurance plans in the data set into four types based on generosity. Polyakova (2016) began by demonstrating substantial adverse selection in the Medicare Plan D market, with more generous contracts attracting individuals with a higher risk score index and with most generous contracts unraveling during the first year of the program. These calculations did not take into account risk adjustment payments that firms received to offset the risk selection; rather they provided a justification for these policies. Like Abaluck and Gruber (2016a), the author presents evidence of switching frictions in this market. In particular, she found that a large majority of consumers enrolled in their last year’s plan and that, despite significant changes in the contract characteristics over time, the distribution of contracts selected by a given cohort reflecting market conditions of the first year, this cohort entered the program and changed little over time.

To quantify the effect of inertia on the degree of adverse selection and welfare, Polyakova developed a structural model of insurance choice where the individual’s (i) utility from choosing plan j in year t is given by


In this equation, pjt denotes premiums, Φjt includes other financial characteristics of the contract, and 1{Default}ijt is equal to one if the individual is enrolled in the last year’s plan.2 Switching friction is captured by the coefficient on the lagged plan choice, γit, which enters the utility of a plan together with a plan’s financial characteristics and brand fixed effects. Parameters γit are allowed to vary with an individual’s demographic characteristics and health risk, similar to other time-varying parameters in this model. The term ϵijt is assumed to follow type 1 extreme value distribution. Identification of the inertia parameters is achieved by specifying time-invariant unobserved heterogeneity in plan preferences and by utilizing data on choices in the first year of consumers’ eligibility for the program and in the year of the program’s inception. The estimated switching friction is substantial: Individuals are willing to give up $1,000 on average to stay in the last year’s plan, with heterogeneity across demographic groups.

Polyakova (2016) also estimated a reduced form plan pricing rule of insurers by regressing the plans’ premiums on several moments of the last year’s costs and the plan’s financial attributes. This pricing rule is used to generate new plans’ premiums in the counterfactual scenarios that change the allocation of consumers to contracts. The equilibrium effects of removing inertia from the plan choice behavior are simulated. The study finds that removing inertia decreases adverse selection and average premiums as consumers switch into cheaper plans. Assuming that switching friction is welfare neutral, removing inertia also increases consumer welfare from better matching of consumers’ preferences to contracts by $455 (23% of annual drug spending). The article is agnostic about the interpretation of this switching friction, however, and there is no attempt to separate consumer inertia into factors such as time and effort cost of search, barriers in acquiring information about the plans, and cognitive limitations in processing this information.

To summarize, a number of studies based on large representative samples from the Medicare population demonstrated that consumers make substantial and costly mistakes in their choices of prescription drug insurance in the market for the Medicare Part D plans. Potential reasons for these mistakes, and in some cases supported by empirical evidence, include a lack of basic understanding of insurance, choice overload due to a large number of options in the choice sets, and informational frictions whereby consumers do not use all available information about plans when making their choices or do not attend to the choice problem itself unless prompted by external factors. Given the advanced age of the population under study (65 years old and above), some of these mistakes can be explained by the decline in cognitive and financial skills that tends to accompany aging (Keane & Thorp, 2016). The next section, “Choice Inconsistencies in Employer-Provided Insurance,” surveys studies that demonstrate that health insurance choices of younger consumers are similarly prone to mistakes and inconsistencies.

Choice Inconsistencies in Employer-Provided Insurance

Several studies have demonstrated that consumers make mistakes in the context of employer-provided health insurance. Sinaiko and Hirth (2011) is one of the first studies to demonstrate that consumers can actively select a health insurance plan that is dominated by other options in the consumers’ choice set. The authors utilized administrative data on employees at the University of Michigan, covering the time period 2002–2004, that included information on plan choices, salary, and other job characteristics. The authors studied choices of employees considering two plans—an HMO A and a POS plan. Both plans had the same premiums and cost-sharing design and covered the same provider network. The POS plan, however, was more generous in that it additionally covered in-network self-referrals to specialists, partially covered out-of-network medical services, and had some other additional benefits. Hence, the HMO A could be considered a dominated plan in this context. The authors showed that during the study period, 39% of new workers and 35% of all workers were enrolled in the dominated HMO A plan. The authors also showed that plan switching was very rare, suggesting substantial inertia (3.3% of the sample switched plans between 2002 and 2004), and while a substantial fraction of switches was from the HMO A to the POS plan (42% of switchers), many workers also switched from the POS into the dominated HMO A plan (20% of switchers). Enrolling into the dominated plan potentially had substantial out-of-pocket costs for those employees who self-referred to specialists and used out-of-network services.

Handel (2013) also documents substantial inertia in the choice of employer-provided health insurance in their setting, and demonstrates that this inertia prevents consumers’ switching out of plans that become dominated over time. The study used detailed data on health insurance plan choices and medical utilization of employees of a large U.S. firm over the period 2004–2009. To reliably identify inertia and to separate it from choice persistence due to unobserved heterogeneity, the authors utilized an exogenous change in the menu of health plans available to the employees: In year to the firm completely changed its menu of plans and encouraged active choice from the new menu by not setting the default option and by carrying out an information campaign.

Another advantage of the data is that the new plans differed from each other only by premiums and cost-sharing characteristics (deductible, co-insurance rate, and maximum out-of-pocket costs) and shared most of the other attributes such as provider characteristics and quality. This setup allowed the author to abstract from consumer preferences for the unobserved plan’s characteristics (e.g., brand loyalty), which complicate insurance choice modeling in other settings.

Identification of inertia relies on comparison of choices the consumers made at the time of the menu change (“the active choice year”) to those in subsequent years when consumers could default into their incumbent plan without taking any action. The author showed that in the year following to, the premiums changed substantially, making some plans financially dominated by other plans for most consumers. However, consistent with the presence of choice inertia, a large fraction of employees did not change their now dominated default plans despite significant potential financial gains from re-optimizing their choice. Furthermore, plan choices of new employees who started in year t1 were more consistent with year t1's plan prices than those of employees with longer tenure who made their initial plan choice in year to. The study also presents evidence of adverse selection into the most generous plan at time to, which did not change substantially at t1 even after the plan prices changed dramatically, further supporting the presence of inertia in consumer choice.

The econometric choice model specifies the following utility of employee i from choosing plan j in period t:


where fijt(OOP) is the consumer’s distribution of prospective out-of-pocket medical costs in plan j, estimated using detailed information on consumers’ medical utilization and diagnoses prior to year to and financial characteristics of the plans; Pijk are plans’ annual premiums, which depend on family income and other characteristics, Wi is the family’s wealth, and ui is the constant absolute risk aversion (CARA) utility function with risk preference parameters that can vary with individual characteristics. Inertia is measured by the effect of the lagged plan choice, ιij,t1, on the probability to choose the same plan in the current year, ceteris paribus, and is allowed to vary with individual characteristics. The employee chooses the plan with the highest Uijt.

The results suggest a substantial inertia in the employee plan choices: Willingness to pay for staying in the last year’s plan is equal to $1,729 for single employees and $2,480 for employees with dependents. Employees in worse health have higher inertia. The average employee in the data set forgoes $2,032 in expected savings by not re-optimizing her choice of the health plan when prices of alternatives in the choice set change.

The author points out that this estimate is too large to rationalize inertia by the explicit search and transaction costs associated with changing the plan and provides a discussion about other potential sources of inertia, such as imperfect information about characteristics of the choice set, which prevents consumers from re-optimizing their plan holdings.

The counterfactual simulations investigate the effects of removing inertia on adverse selection and welfare. In these simulations, the plans are repriced according to the changes in risk profiles of their enrollees due to switches using the average cost plus a markup rule. A priori the direction of the effect of reducing inertia on selection is ambiguous. In theory, lower inertia can reduce adverse selection over time if inertia and risk are positively correlated. The results suggest that reducing inertia increases adverse selection in the setting of this study. In particular, as reduced inertia made healthier consumers switch out of a generous plan, this plan was virtually eliminated from the market over time in the adverse selection “death spiral” of growing costs and premiums. So, despite that decreasing inertia improved matches between the characteristics of the plans and needs of the consumers, the net effect on welfare was negative due to the increase in adverse selection. For example, reducing inertia to 25% of its estimated value decreases welfare by $115 per person per year for an average individual (7.7% of average annual premium) with substantial distributional consequences. Those who switched plans as a result of the policy actually experienced a welfare gain of $186 per person per year (12% of the average premium), while those who did not switch lost $442 (29.4%) per year. The main reason for the decrease in welfare among stayers was the increase in premiums in their chosen plan.

The study by Handel and Kolstad (2015) is one of the first that points out inaccurate beliefs about attributes of health plans as an important reason for poor quality of consumer choices of employer-provided health insurance. The distinctive feature of this study is that it uses rich administrative data on health insurance plan choices and medical utilization of consumers that are linked to a survey about their knowledge of health plans’ attributes and perceived hassle costs of managing the plans. The authors use this rarely available information on subjective beliefs to estimate the demand frictions due to information constraints and perceived hassle costs and to quantify how incorporating these frictions into the health insurance choice model changes estimated risk preference parameters.

The data came from a large U.S. employer and included information on health plan choice sets, actual choices, and medical utilization for about 160,000 employees and their families. During the time period 2009–2012, the employees chose between two health insurance options, the PPO and HDHP. The plans had the same non-financial attributes (e.g., provider network and treatments covered) but had different financial attributes (i.e., premiums, deductibles, co-insurance rates, etc.). In particular, the PPO option had a simpler design which did not require deductibles, co-insurance, or out-of-pocket maximums for in-network services. In contrast, the HDHP plan featured a substantial deductible which varied with family characteristics, different co-insurance rates for in-network and out-of-network providers, and an out-of-pocket maximum that varied with family characteristics. Hence, the HDHP plan featured more risky healthcare expenditure, and possibly was more difficult for consumers to comprehend. The employees did not have to pay premiums for either plan and received a subsidy from the employer for the HDHP plan. The authors showed that the financial gains from selecting the HDHP relative to the PPO depended on the total medical expenditures, and that there was a unique level of expenditure below which the PPO plan financially dominated the HDHP plan. Under some realistic assumptions, the authors estimated that about 60% of employees would be ex-post better off in the HDHP plan than in the PPO plan. Despite these potential financial benefits, fewer than 20% of employees chose the HDHP plan during the study period. The authors showed that the standard health insurance choice model without informational and hassle cost frictions would require implausibly large levels of risk aversion to rationalize these choices.

The analysis focused on years 2011–2012 when the administrative data was linked with individual surveys containing close to 30 multiple-choice questions eliciting consumers’ knowledge of financial and non-financial characteristics of the HDHP plan, as well as beliefs about time costs of managing this plan, subjective preferences for hassle, and overall satisfaction with their current plan. The responses revealed that enrollees of both plans had limited knowledge of financial and non-financial properties of the HDHP, with the PPO enrollees significantly less likely to answer correctly compared to the HDHP enrollees. Strikingly, only 49% of the HDHP enrollees and 32% of the PPO enrollees were aware that the two plans provided access to the same network of physicians. Also, the PPO enrollees reported significantly higher expected hassle costs of the HDHP plan compared to the HDHP enrollees. The actual plan choices of consumers were consistent with these beliefs—consumers who correctly believed that both plans offered the same network of providers were significantly more likely to choose the HDHP. The propensity to choose the HDHP declined with the perceived hassle costs of this plan.

These findings suggest that subjective beliefs about characteristics of health plans are an important determinant of consumers’ choices. A model of health insurance plan choice that does not explicitly model these factors will attribute choice heterogeneity due to demand frictions to differences in fundamental risk preferences. To quantify this bias, the authors estimated the insurance choice model with and without controlling for consumers’ perceptions and beliefs, and analyzed the consequences of omitting these subjective factors for the estimated risk preferences and welfare.

The empirical approach was very similar to that in Handel (2013). The two baseline models (with and without inertia) are similar to the model in Equation 3.3 The full models additionally include variables measuring informational frictions and hassle costs derived from the survey responses as arguments shifting the utility, ui.

The results suggest that subjective beliefs affect willingness to pay for the plans and are important determinants of consumer choices. For example, those who believed that the PPO plan gave access to a wider network of providers valued the HDHP by $2,326 less than someone who had correct beliefs. Similarly, for each hour of perceived HDHP hassle time, the value of the HDHP decreased by $138.70 for consumers who strongly disliked hassle costs, and by $127.87 for consumers who were more tolerant of such costs. At the aggregate, the frictions reduced consumer willingness to pay for the HDHP plan by $1,787. Accounting for the information frictions and hassle costs substantially reduced the estimated mean and variance of risk aversion.

Welfare analysis focused on the policy change that eliminated the PPO and forced all employees to switch into the HDHP plan. This policy was actually implemented by the firm in the years following the analysis. Because the HDHP plan provided less risk protection than the PPO, reliable estimates of consumers’ risk aversion are crucial for understanding welfare consequences of this change. As discussed in “Demand for Health Insurance: Theory,” the welfare analysis assumed that the demand frictions and inertia were welfare neutral and did not affect the ex-post welfare experienced in the HDHP plan, which primarily depended on risk preferences and the distribution of risk.

The welfare analysis reveals that accounting for demand frictions substantially changes conclusions about the welfare impact of the policy. In particular, the baseline model without inertia and demand frictions that gives rise to the highest estimated risk aversion implies that the policy would decrease welfare by $1,237, on average. In the baseline model with inertia, the mean welfare loss is $874. The full model accounting for all frictions predicts an even smaller decrease in consumer surplus of $788. The welfare analysis undertaken in the article did not allow for the HDHP repricing after the forced switch, because HDHP price remained the same after the actual policy change implemented by the firm. The policy with endogenous repricing could have had different welfare consequences.

Bhargawa et al. (2017) identify consumers’ lack of basic health insurance competence as another important reason for the mistakes. The study used rich administrative data for the period 2010–2012 on health insurance plan choices, medical expenditure, and socio-demographic characteristics of about 24,000 employees of a large Fortune 500 U.S. firm. In the year 2010, the firm introduced a new health insurance program in which employees could “build” a plan from a standardized menu of 48 options that differed only on premiums and four cost-sharing dimensions (i.e., deductibles, maximum out-of-pocket spending above the deductible, co-insurance rates, and copayments for doctors’ visits). The plans were identical in all other dimensions (e.g., provider coverage, quality).

The authors demonstrated that of the 36 low deductible plans in the menu, 35 were financially dominated by a high-deductible plan. For example, plan A with a deductible of $500 and premium of $1,568 is financially dominated by an otherwise identical plan B with a deductible of $1,000 and premium of $930 because plan A results in higher out-of-pocket expenditure than plan B, with absolute certainty.4 In other words, for any level of total medical expenditure, the out-of-pocket cost schedule associated with plan A is above that of plan B.

When the new program was introduced, the employees were encouraged by the firm to make active plan choices. The employees who did not make the active choice were automatically defaulted into one of the non-dominated plans, but the authors show that a large majority of employees (98%) actively chose their new plan. Despite this active choice environment, 61% of employees chose a nominally dominated plan, while 55% of employees chose a plan that was financially dominated after adjusting plan premiums by the employee’s marginal tax rate. Lower-salaried and sicker employees were more likely to make dominated choices. These inefficient choices came at financial costs—on average employees in dominated plans would have saved $372 per year (24% of their original plan premiums) virtually without risk of losing money by switching to an otherwise identical plan with a high ($1,000) deductible. These conclusions are largely unaffected by different assumptions about consumers’ risk aversion. The authors also showed that voluntary plan switches in the following year did not substantially improve the quality of consumer choices.

The authors then tested if the demand frictions such as (a) choice overload due to menu complexity (e.g., Iyengar & Kamenica, 2010; Frank & Lamiraud, 2009), (b) informed preferences for the low-deductible plans, or (c) low insurance competence could explain the consumers’ choices of dominated plans. To this end, they carried out two online experiments in which subjects with similar socio-demographic profiles as those of employees in the firm’s data were recruited from the Qualtrics and Amazon Mechanical Turk platforms and asked to make a series of hypothetical health insurance choices from the menu and within the enrollment interface similar to those encountered by employees in the earlier analysis.

In the first experiment, designed to measure the effect of choice complexity, the subjects were presented with different choice menus that varied by size (12 plans vs. 4 plans), number of attributes (two vs. four), and choice interface (sequential presentation of plan characteristics mimicking the actual choice interface, and a simultaneous display of all options and prices in a single table). The results suggested that while the simultaneous display somewhat reduced the probability of selecting a dominated plan, other treatments reducing choice complexity did not lead to further improvement of consumer choices. In fact, 66% of the subjects chose financially dominated low-deductible plans under the simplest choice condition featuring four plans that differed only in terms of deductibles and premiums.

Since the choice complexity did not explain choice inconsistencies in the first experiment, the authors carried out the second experiment to investigate whether these inconsistencies stemmed from nonstandard preferences or rather were a consequence of consumers’ poor understanding of insurance. To isolate the effect of insurance competence, they compared choices of subjects that were randomly assigned to choice menus varying in the clarity of presentation of financial consequences of plan choices (e.g., out-of-pocket costs under good and bad health realizations) and insurance terminology. The subjects were also given assessments of their health insurance literacy and comprehension of insurance vocabulary.

The results suggested a striking decrease in the probability of making a dominated choice among the subjects in the high-clarity treatment group (to 18% from the 48% at the baseline). The subjects were significantly less likely to make a dominated choice when the financial implications of plan choices were clearly explained to them. The results also revealed widespread deficits in understanding of insurance among the subjects, with 30% to 65% of respondents answering basic insurance competence questions incorrectly. The propensity to choose a dominated plan was very strongly correlated with insurance incompetence.

The authors argue that these patterns point to the lack of insurance competence as the key determinant of dominated choices in their experimental and field data. If these choice inconsistencies were driven by nonstandard informed preferences that favored low deductibles, then additional information explaining insurance terminology and financial trade-offs of different options would not have affected the choices. In the discussion, the authors caution against expansion of choice in health insurance, assuage concerns about unraveling of health insurance markets due to adverse selection, and caution against estimating risk preferences using models that do not account for informational and behavioral demand frictions from the data where choices are known to be affected by them.

Concluding Remarks

In the second decade of the 21st century, several studies have shown that consumers leave significant amounts of money on the table by making poor choices of health insurance. A variety of mistakes in the health insurance choices have been identified, such as valuation of financial plan attributes beyond their impact on out-of-pocket costs, active choice of financially dominated options, and inertia, which exacerbates the initial mistakes and leads to additional losses. These mistakes cost hundreds and even thousands of dollars to consumers and tend to be more prevalent among those with lower income and educational attainment and those in worse health. The research has shown that inattention to plans’ attributes, inaccurate beliefs and perceptions about plans’ characteristics, and a lack of basic insurance competence go a long way toward explaining the incidence of mistakes. It pointed out significant difficulties that consumers experience with acquiring and processing information necessary for making efficient choices. There is also evidence that consumers’ health insurance choices are sensitive to how price differences between insurance plans are presented (i.e., in absolute or percentage terms), pointing out behavioral anomalies such as “relative thinking” in consumer behavior in this market (e.g., Douven, van der Heijden, McGuire, & Schut, 2017; Schmitz & Ziebarth, 2017).

In view of the growing evidence of consumers’ confusion in the markets for health insurance and other complex financial products, many prominent economists and policy makers have been advocating policies that simplify or facilitate consumer choice in these markets. Examples of such policies in the health insurance context include government regulation of financial and non-financial characteristics of health insurance options offered by private insurers (e.g., minimum cost-sharing requirements, or minimum service coverage), setting smart defaults, framing consumers’ choices through web design and customized information, standardization of health plans, and educational interventions (Handel & Schwartzstein, 2018).

Although the extent to which these policies can fully eliminate confusion is not yet known, there is evidence that some of them do improve consumers’ choices. Several studies surveyed in this article demonstrate that customized information about out-of-pocket costs and other features of different plans helped consumers choose more cost-effective options (e.g., Kling et al., 2012; Bhargava et al., 2017). Ericson and Starc (2016) showed that the policy change that standardized cost-sharing parameters of plans across insurers and changed the presentation of information to consumers in the Massachusetts Health Insurance Exchange improved choices of health insurance plans, as consumers put more weight on cost-sharing properties of alternative plans post-policy. However, Abaluck and Gruber (2016b) showed that in their setting with employer-provided health insurance in school districts in Oregon, policy interventions such as promotion of active choice and the introduction of decision support software did little to improve consumers’ choices, largely because consumers did not follow the recommendations. However, reducing the number of options in consumers’ choice sets led to welfare improvement. The improvement happened not because the policy reduced choice overload, but because in their setting larger choice sets included more poorly designed plans.

While the potential of confusion-reducing policies to increase consumers’ welfare by improving matching between plans’ characteristics and consumers’ preferences and needs is clear, several studies reviewed in this article cautioned that such policies can also exacerbate adverse selection as sorting of consumers into plans with different generosity based on risk improves. The evidence to date about the net welfare effects of policies reducing confusion is mixed. For example, Handel (2013), Handel, Kolstad, and Spinnewijn (in press), and Ketcham, Kuminoff, and Powers (2016) found that policies that eliminated consumers’ inertia would reduce welfare due to increased adverse selection, while Polyakova (2016) showed that such policies could decrease adverse selection and improve welfare in the Medicare Part D setting. The potential of confusion-reducing policies to exacerbate adverse selection suggests that these policies may have to be implemented as part of the mix with risk-adjustment policies (Handel et al., in press).

Equilibrium welfare analysis of policies reducing consumer confusion in the market for private health insurance is incomplete without understanding of the supply-side responses to these policies. For example, it is well-known that in theory consumer inertia creates incentives for insurers to follow the “invest-then-harvest” pricing strategy where firms offer low prices initially to attract customers (“invest”), and subsequently increase prices as the insureds become locked into their contracts and lose sensitivity to price (Farrell & Klemperer, 2007). Ericson (2014) demonstrates that evolution of prices in the Medicare Part D market may be consistent with this strategy. Consumers’ lack of insurance competence may also create incentives for firms to compete on product features that are salient to consumers rather than on cost-sharing and quality. Therefore, the policies that decrease confusion are likely to change firms’ behavior as well. As of 2019, the empirical equilibrium analyses of policies that reduce demand frictions in the health insurance market did not model how firms respond to changes in consumers’ sophistication, but this area is an exciting one for future research.

Additional Studies

In addition to field studies surveyed in this article, the second decade of the 21st century has seen a growth of experimental studies of the effects of choice complexity, choice architecture, and insurance literacy on the quality of consumer choices in the health insurance domain (e.g., Barnes, Hanoch, & Rice, 2014; Johnson, Hassin, Baker, Bajger, & Treuer, 2013; Kairies-Schwarz, Kokot, Vomhof, & Weßling, 2017; Kettlewell, in press; Schram & Sonnemans, 2011). Departures from rational behavior have also been pointed out in the context of healthcare choices as consumers demand treatments according to subjective perceptions of their value and not according to their objective health benefits (Baicker, Mullainathan, & Schwartzstein, 2015), and as consumers adjust their demand of various health treatments suboptimally in response to changes in cost-sharing characteristics of their health plans (Brot-Goldberg, Chandra, Handel, & Kolstad, 2017). A body of research that developed in the first two decades of the 21st century presents evidence that insurance choices in various domains, in addition to standard risk aversion, also exhibit features consistent with nonstandard risk preferences (Barseghyan, Molinari, O’Donoghue, & Teitelbaum, 2013; Barseghyan, Molinari, O’Donoghue, & Teitelbaum, 2018). A survey of behavioral economics contributions to welfare analysis of public policies more generally is provided in Bernheim and Taubinsky (2018).


Abaluck, J., & Gruber, J. (2011). Choice inconsistencies among the elderly: Evidence from plan choice in the Medicare Part D program. American Economic Review, 101(4), 1180–1210.Find this resource:

Abaluck, J., & Gruber, J. (2016a). Evolving choice inconsistencies in choice of prescription drug insurance. American Economic Review, 106(8), 2145–2184.Find this resource:

Abaluck, J., & Gruber, J. (2016b). Improving the quality of choices in health insurance markets. NBER Working Paper No. 22917, Cambridge, MA.Find this resource:

Baicker, K., Mullainathan, S., & Schwartzstein, J. (2015). Behavioral hazard in health insurance. Quarterly Journal of Economics, 130, 1623–1667.Find this resource:

Barnes, A. J., Hanoch, Y., & Rice, T. (2014). Determinants of coverage decisions in health insurance marketplaces: Consumers’ decision-making abilities and the amount of information in their choice environment. Health Services Research, 50, 58–80. Advance online publication.Find this resource:

Barseghyan, L., Molinari F., O’Donoghue, T., & Teitelbaum J. C. (2013). The nature of risk preferences: Evidence from insurance choices. American Economic Review, 103(6), 2499–2529.Find this resource:

Barseghyan, L., Molinari, F., O’Donoghue, T., & Teitelbaum J. C. (2018). Estimating risk preferences in the field. Journal of Economic Literature, 56(2), 501–564.Find this resource:

Bernheim, B. D., & Taubinsky, D. (2018). Behavioral public economics. NBER Working Paper No. 24828, Cambridge, MA.Find this resource:

Bhargava, S., Loewenstein, G., & Sydnor, J. (2017). Do individuals make sensible health insurance decisions? Evidence from a menu with dominated options. Quarterly Journal of Economics, 132(3), 1319–1372.Find this resource:

Brot-Goldberg, Z. C., Chandra, A., Handel, B. R., & Kolstad, J. T. (2017). What does a deductible do? The impact of cost-sharing on health care prices, quantities, and spending dynamics. Quarterly Journal of Economics, 132, 1261–1318.Find this resource:

Bundorf, K. M., Levin, J., & Mahoney, N. (2012). Pricing and welfare in health plan choice. American Economic Review, 102(7), 3214–3248.Find this resource:

Cafferata, G. (1984). Knowledge of their health insurance coverage by the elderly. Medical Care, 22(9), 838847.Find this resource:

Cohen A., & Einav, L. (2007). Estimating risk preferences from deductible choice. American Economic Review, 97, 745–788.Find this resource:

Cutler, D. M., & Zeckhauser, R. J. (2000). The anatomy of health insurance. In A. J. Culyer & J. P. Newhouse (Eds.), Handbook of health economics (Vol. 1, Part A, pp. 563–643). Amsterdam, The Netherlands: Elsevier.Find this resource:

Cutler, D. M., & Zeckhauser, R. (2004). Extending the theory to meet the practice of insurance. In R. Litan & R. Herring, (Eds.), Brookings-Wharton Papers on Financial Services. Washington, DC: Brookings Institution.Find this resource:

Davidson, B., Sofaer, S., & Gertler, P. (1992). Consumer demand and biased selection in the demand for coverage supplementing Medicare. Social Science and Medicine, 34(9), 10231034.Find this resource:

Douven, R., van der Heijden, R., McGuire, T., & Schut, F. T. (2017). Premium levels and demand response in health insurance: Relative thinking and zero-price effects. NBER Working Paper No. 23846, Cambridge, MA.Find this resource:

Einav, L., Finkelstein, A., & Levin, J. (2010). Beyond testing: Empirical models insurance markets. Annual Review of Economics, 2010(2), 311–336.Find this resource:

Ericson, K. M. (2014). Consumer inertia and firm pricing in the Medicare Part D prescription drug insurance exchange. American Economic Journal: Economic Policy, 6, 38–64.Find this resource:

Ericson, K. M., & Starc, A. (2016). How product standardization affects choice: Evidence from the Massachusetts Health Insurance Exchange. Journal of Health Economics, 50, 71–85.Find this resource:

Ericson, K. M., & Sydnor, J. (2017). The questionable value of having a choice of levels of health insurance coverage. Journal of Economic Perspectives, 31(4), 51–72.Find this resource:

Fang, H., Keane, M. P., & Silverman, D. (2008). Sources of advantageous selection: Evidence from the Medigap insurance market. Journal of Political Economy, 116, 303–349.Find this resource:

Farrell, J., & Klemperer, P. (2007). Coordination and lock-in: Competition with switching costs and network effects. In M. Armstrong & R. H. Porter (Eds.), Handbook of industrial organization (Vol. 3, pp. 1967–2072). Amsterdam, The Netherlands: Elsevier.Find this resource:

Frank, R. G., & Lamiraud, K. (2009). Choice, price competition and complexity in markets for health insurance. Journal of Economic Behavior & Organization, 71, 550–562.Find this resource:

Frank, R. G., & Zeckhauser, R. J. (2009). Health insurance exchanges: Making the markets work. New England Journal of Medicine, 361, 1135–1137.Find this resource:

Gruber, J. (2017). Delivering public health insurance through private plan choice in the United States. Journal of Economic Perspectives, 31(4), 3–22.Find this resource:

Handel, B. (2013). Adverse selection and inertia in health insurance markets: When nudging hurts. American Economic Review, 103, 2643–2682.Find this resource:

Handel, B., & Kolstad, J. (2015). Health insurance for “humans”: Information frictions, plan choice and consumer welfare. American Economic Review, 105, 2449–2500.Find this resource:

Handel, B., & Schwartzstein, J. (2018). Frictions or mental gaps: What's behind the information we (don't) use and when do we care? Journal of Economic Perspectives, 32(1), 155–178.Find this resource:

Handel, B., Kolstad, J., & Spinnewijn, J. (in press). Information frictions and adverse selection: Policy interventions in health insurance markets. Review of Economics and Statistics.Find this resource:

Harris, K. M., & Keane, M. P. (1998). 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.Find this resource:

Heiss, F., Leive, A., McFadden, D., & Winter, J. (2013). Plan selection in Medicare Part D: Evidence from administrative data. Journal of Health Economics, 32(6), 1325–1344.Find this resource:

Heiss, F., McFadden, D., & Winter, J. K. (2010). Mind the gap! Consumer perceptions and choices of Medicare Part D prescription drug plans. In D. A. Wise (Ed.), Research findings in the economics of aging. Chicago, IL: University of Chicago Press.Find this resource:

Ho, K., Hogan, J., & Morton, F. S. (2017). The impact of consumer inattention on insurer pricing in the Medicare Part D program. RAND Journal of Economics, 48(4), 877–905.Find this resource:

Iyengar, S., & Kamenica, E. (2010). Choice proliferation, simplicity seeking, and asset allocation. Journal of Public Economics, 94, 530–539.Find this resource:

Johnson, E. J., Hassin, R., Baker, T., Bajger, A. T., & Treuer, G. (2013). Can consumers make affordable care affordable? The value of choice architecture. PLOS ONE, 8, e81521–e81521.Find this resource:

Kairies-Schwarz, N., Kokot, J., Vomhof, M., & Weßling, J. (2017). Health insurance choice and risk preferences under cumulative prospect theory: An experiment. Journal of Economic Behavior and Organization, 137, 374–397.Find this resource:

Keane, M. P., & Stavrunova, O. (2016). Adverse selection, moral hazard and the demand for medigap insurance. Journal of Econometrics, 190(1), 62–78.Find this resource:

Keane, M. P., & Thorp, S. (2016). Complex decision making: The roles of cognitive limitations, cognitive decline and ageing. In J. Piggott & A. Woodland (Eds.), Handbook of the economics of population aging (Vol. 1, pp. 661–709). Amsterdam, The Netherlands: Elsevier.Find this resource:

Ketcham J. D., Kuminoff, N. V., & Powers, C. A. (2016). Estimating the heterogeneous welfare effects of choice architecture: An application to the Medicare prescription drug insurance market. NBER Working Paper No. 22732, Cambridge, MA.Find this resource:

Ketcham, J. D., Lucarelli, C., Miravete, E. J., & Roebuck, M. C. (2012). Sinking, swimming, or learning to swim in Medicare Part D. American Economic Review, 102(6), 2639–2673.Find this resource:

Kettlewell, N. (in press). Policy choice and product bundling in a complicated health insurance market: Do people get it right? Journal of Human Resources.Find this resource:

Kling, J. R., Mullainathan, S., Shafir, E., Vermeulen, L. C., & Wrobel, M. V. (2012). Comparison friction: Experimental evidence from Medicare drug plans. Quarterly Journal of Economics, 127(1), 199–235.Find this resource:

Liebman, J., & Zeckhauser, R. (2008). Simple humans, complex insurance, subtle subsidies. NBER Working Paper 14330, Cambridge, MA.Find this resource:

McCall, N., Rice, T., & Sangl, T. (1986). Consumer knowledge of Medicare and supplemental health insurance benefits. Health Services Research, 20(6), 633657.Find this resource:

McFadden, D. (2006). Free markets and fettered consumers. American Economic Review, 96, 3–29.Find this resource:

McGuire, T. G. (2012). Demand for health insurance. In M. Pauly, T. G. McGuire, & P. Barros (Eds.), Handbook of health economics (Vol. 2). Amsterdam, The Netherlands: Elsevier.Find this resource:

Polyakova, M. (2016). Regulation of insurance with adverse selection and switching costs: Evidence from Medicare Part D. American Economic Journal: Applied Economics, 8(3), 165–195.Find this resource:

Schmitz, H., & Ziebarth, N. R. (2017). Does price framing affect the consumer price sensitivity of health plan choice? Journal of Human Resources, 52(1), 88–127.Find this resource:

Schram, A., & Sonnemans, J. (2011). How individuals choose health insurance: An experimental analysis. European Economic Review, 55, 799–819.Find this resource:

Sinaiko, A. D., & Hirth, R. A. (2011). Consumers, health insurance and dominated choices. Journal of Health Economics, 30, 450–457.Find this resource:

Schmueli, A., Stam, P., Wasem, J., & Trottmann, M. (2015). Managed care in four managed competition OECD health systems. Health Policy, 119, 860–873.Find this resource:

Spinnewijn, J. (2017). Heterogeneity, demand for insurance and adverse selection. American Economic Journal: Economic Policy, 9(1), 308–343.Find this resource:

Winter, J., Balza, R., Caro, F., Heiss, F., Jun, B., Matzkin, R., & McFadden, D. (2006). Medicare prescription drug coverage: Consumer information and preferences. Proceedings of the National Academy of Sciences USA, 103, 7929–7934.Find this resource:


(1.) It is also possible that consumers have full information about attributes and costs of plans, but different preferences; that is, v(wi,e,pj,λ^j,,θjξi)u(wipje,λ^j,ξi). For example, they can attach different valuations to costs associated with a plan’s premiums and out-of-pocket expenditures. This can happen if consumers are liquidity constrained and find it difficult to finance health care losses early in the year (before the out-of-pocket expenditures exceed the deductible). If premiums are paid smoothly over the year, the consumers will attach a lower disutility to the costs associated with a plan’s premiums than to equal out-of-pocket costs (Bhargava et al., 2017).

(2.) In the case of renewal or consolidation of last year’s plan, 1{Default}ijt is equal to 1 if the individual is enrolled in the default plan determined by the CMS.

(3.) The first baseline model does not include the inertia term ιij,t1.

(4.) In particular, plan A has a deductible that is $500 lower than that in plan B, but features a premium that is $630 higher.