- 1 Although Arrow introduced the term, Pauly’s analysis of moral hazard is largely responsible for emb (...)
1This is a history of the concept of moral hazard in health insurance and of the controversies associated with it among health economists. The term “moral hazard” was used for the first time in the context of health insurance by K.J. Arrow in his seminal 1963 article on medical care to characterize the fact that the insured use more health care services to treat a given illness than the uninsured (Arrow, 1963). He borrowed the expression from an industry for which he had worked in his youth (Finkelstein et al., 2016) to describe something he saw as a “practical limitation on the use of insurance” (Arrow, 1963, 961). While moral hazard in the insurance literature is taken to mean deviation from “correct” behaviour (Buchanan, 1964, 22, cited in Pauly, 1968, 535), or “failure to uphold the accepted moral qualities,” (Faulkner, 1960, 327, also cited in Pauly, 1968, 535), Pauly (1968) subsequently argued that the application of this concept to health insurance was a misnomer. Pauly stated that, under an insurance contract that reduces the price at the point of use, there was nothing unethical or immoral in the response by an insured individual to use more services than when uninsured. Rather, the insured individual was simply reacting as any rational individual would (from any standard economics textbook) faced with a change in the price of a commodity for which they had tastes that could be translated into demand (conditional on their level of income). Pauly (1968, 535) opposed the value-laden view of the insurance industry in favour of what he saw as value neutral economic analysis:1 “It is surprising that very little economic analysis seems to have been applied here”.
2Moral hazard (in health or otherwise) divides into ex ante and ex post moral hazard, the former being concerned with changes in the probability of the event, and the latter with changes in the cost of the event conditional on the event occurring. Arrow’s and Pauly’s articles in the 1960s focused only on ex post moral hazard, and here we address the history of ex post moral hazard in health insurance: coverage-induced variation in average spending on health care following illness or injury and its welfare implications. We exclude consideration of ex ante moral hazard because it was not the topic of a controversy among health economists as ex post moral hazard was and is often seen as a relatively minor issue in health: the true cost of an illness is the pain, suffering, assault on personal integrity, and risk of dying, which no insurance contract covers and, as a result, it is hard to believe that individuals will take risks with their health because the financial risk is covered (Cutler and Zeckhauser, 2000). Further, we have little empirical evidence regarding ex ante moral hazard: in their review of consumer incentives in health care, Zweifel and Manning (2000) identified only four empirical studies and concluded that empirical evidence is “limited” and measured on heterogeneous populations.
- 2 Interestingly, although Leaver (2015) characterizes Pauly’s mainstream approach as neo-classical an (...)
3No history of the concept of moral hazard in health insurance, and of its consequences on policy recommendations made by health economists, has ever been written, and existing histories of moral hazard in economic thought in general pay scant attention to moral hazard in relation to health insurance (Baker, 1996; Rowell and Connelly, 2012; Loubergé, 2013; Leaver, 2015; and Latsis and Repapis, 2015). Loubergé (2013) singles out Arrow 1963 as a major contribution to the history of economic modelling of risk, but his discussion of moral hazard excludes Pauly’s contribution. His account puts more emphasis on ex ante moral hazard and restricts the definition of ex post moral hazard to fraudulent reporting, such as exaggerations of the value of a good or the severity of the damage. The other four histories we identified include references to and discussions of Arrow and Pauly’s contributions but limit their discussion of moral hazard in health to the initial controversy between Arrow and Pauly in the 1960s, and conclude that Pauly’s views prevailed. These histories cite this (often disapprovingly) as an example of the triumph of a de-contextualized, formalist version of welfare economics over a more institutionally and socially conscious one represented by Arrow.2
4That these general reviews pay scant attention to moral hazard in health is telling: health economists have developed a specific literature on moral hazard that has not really communicated with the literature on moral hazard in mainstream economics. Holmström (1979), perhaps the most cited article on moral hazard in mainstream economics and a starting point of the literature on moral hazard in economics (Leaver, 2015) does not mention Arrow (1963) or Pauly (1968), and his only reference to ex post moral hazard in health is Zeckhauser (1970). Moreover, even though Holmström’s paper has been cited more than 9,000 times, a Google Scholar search indicates only a handful of citations by health economists, then only in reference to provider payment and physician agency, never in reference to moral hazard in health insurance.
5We build on these reviews to assess the Arrow-Pauly debate regarding moral hazard in health insurance, but propose a different interpretation than the predominant story of these reviews and current textbooks. Arrow and Pauly certainly disagreed at a deep epistemological level about moral hazard, but relatedly, they also had strikingly different views of the demand for health care. For health economics, this latter difference has been one of the most enduring aspects of the controversy, a difference that importantly framed subsequent debates on moral hazard in health and anticipated the emergence of a specific branch of economics devoted to health. We argue that health economics is not merely an application of the economist’s tools to the specific area of health or medical care, but is a distinct intellectual tradition that discusses specific issues (such as the nature of demand for health care or demand for health insurance) and develops specific tools to address these issues.
- 3 This is not meant to be an exhaustive list of all textbooks of health economics published in Englis (...)
6This history will show two main divides in the conceptions of moral hazard in health insurance held by health economists, one on the nature of demand for health care—is moral hazard immoral?—and one on the nature of demand for health insurance—is moral hazard a hazard? The first of these followed the initial Arrow-Pauly discussion in the 1960s. There is the perception today among many health economists of a consensus regarding moral hazard in health insurance: Moral hazard is presented in recent textbooks using a standard welfare analysis of the effect of a price subsidy on utilization of health care, as if Pauly (1968) had closed the controversy. Even acknowledging that textbooks tend to de-emphasize controversies and present a unified framework that all students should learn, it is remarkable that, of nine textbooks of health economics published since 19903 (Folland, Goodman and Stano, 2013; Hurley, 2010; Jacobs, 1991; Johnson-Lans, 2006; McPake, Normand and Smith, 2013; Phelps, 1992; Santerre and Neun, 2007; Sloan and Hsieh, 2012, and Zweifel, Breyer and Kifman, 2009), only three (Folland, Goodman and Stano, Hurley, and Santerre and Nun) organize the section on moral hazard around debates and critiques, starting from the “conventional” presentation due to Pauly and then detailing the major conceptual critiques of this mainstream presentation, as well as what they mean in terms of our understanding of the demand for health care. All other textbooks simply recount the mainstream (Pauly) story and present some of the criticisms as potential empirical extensions of the model. Consistent with that perception, Cutler and Zeckhauser (2000) write that Arrow (1984) later conceded the point and agreed to drop the expression “moral hazard” (not only in the field of medical insurance but more broadly in any agency relationship) in favour of value-neutral and flatly descriptive “hidden action” (an action of the agent that the principal cannot observe and that affects the probability distribution of outcomes for the principal).
- 4 “I will call the two types of principal-agent problems hidden-action and hidden-information, respec (...)
- 5 As Rowell and Connelly (2012) point out, this is precisely how the word “moral” was used in the 18t (...)
7However, the concept of moral hazard in health insurance has been contentious from its inception in 1963, and its meaning and policy implications have been continuously disputed up to the present. In fact, Arrow (1984) never states that “hidden action” is preferable to “moral hazard,” only that it applies more broadly.4 The Arrow tradition of seeing moral hazard as a “moral” issue, not so much in the ethical as in the social sense (moral reflecting “mores” the social norms of a society5), has been alive among health economists and has provided the basis for an extra-welfarist interpretation of the demand for health care.
8If the debate on the moral (or not) character of moral hazard has dominated within health economics, another aspect has emerged more recently questioning moral hazard as a “hazard.” In the late 1990s, John Nyman questioned Arrow’s (1963) statement (accepted by most health economists until Nyman) that moral hazard was a “practical limitation” of health insurance (or Pauly’s statement that it was a welfare loss of health insurance) and he stated instead that moral hazard was beneficial because individuals wanted income to be transferred to them when they are sick, in order to spend more on health care. Whereas the debate on the moral character of moral hazard derives from differing perceptions of the nature of the demand for health care, the question of the hazardous nature of moral hazard has opened a debate on the nature of the demand for health insurance.
9The paper is organized as follows: Section 2 will present the arguments on moral hazard in health insurance as they relate to different conceptualizations of the nature of demand for health care and section 3 details the more recent controversy around the nature of demand for health insurance and its consequences on the definition and welfare implications of moral hazard in health insurance.
10We start with accounts of moral hazard as told in the literature, and we will then propose our own reading of the three papers constituting the seminal controversy (Arrow, 1963, Pauly, 1968, and Arrow, 1968), using Pauly (1970; 1978; 1986) as ancillary sources, as they draw out the policy implications that Pauly derives from his 1968 conception of moral hazard. Our reading differs markedly from previous literature and histories on a matter of importance for health economics: we argue that in drawing welfare implications, Arrow emphasized supply-side issues whereas Pauly emphasized demand-side issues. As will be shown below, past histories underscore the role of social and moral institutions, such as trust, in Arrow’s argument, and Pauly’s counter-argument that no theory can be built on such vague and soft concepts. These histories attribute this difference to an epistemological (or even ontological) divide between the two. We agree that such a divide exists, but argue that the main pathway from these fundamental views of economics to the different conceptions of the role of moral hazard in health insurance lies in different interpretations of the relationship between price and utilization of health care, different interpretations not fully attributed to epistemological differences but, rather, to divergent views of how the markets for health care and health insurance work.
11Among the accounts of the controversy, two are most useful for our purpose: Baker (1996), and Latsis and Repapis (2015). Dembe and Boden (2000), another interesting account, provides only a short summary of the seminal articles and its narrative is almost the same as Baker (1996). According to Baker (1996), Arrow recognized that moral hazard was a potential problem for health insurance, including insurance provided by the government (which Arrow supported, Baker, 1996), and Arrow also characterized it as a problem of rational response to incentives, not morality or fraud. However, Baker then accepts Pauly’s contrary claim that Arrow provided a non-economically orthodox definition of moral hazard that places too much emphasis on emotions and morality. Baker then uses Arrow’s response (particularly the example of Judas Iscariot’s action as rational and profitable, but immoral) as confirmation that Arrow stands on the ethical side of the definition of moral hazard whereas Pauly stands on the rational side. Baker concludes that Arrow’s conception was not able to overcome its inherent contradiction: being an economist, he had no theory of how moral principles can be internalized (what Baker calls a concept of “character”) and, as a result, “[i]n the thirty years since their exchange, Pauly's criticism, and not Arrow's response, has had the greater influence. Tellingly, the subsequent economics literature (including Arrow's own contributions) exclusively addresses external incentives, not ‘internalized moral principles.’” (Baker, 1996, 269) Baker is certainly right to state that conventional wisdom among economists (and many health economists) is to see moral hazard as a purely rational reaction to a change in prices, not as a behavior that should be controlled through social institutions and norms. Baker sees this as a failure of the economics profession, resulting from its pro-market and anti-regulation bias.
- 6 “A deeper philosophical divide separates the two authors, despite their reliance on a common techni (...)
12Latsis and Repapis (2015) tell a slightly different story, but the main picture remains the same: Arrow and Pauly differ both ideologically (Arrow supports public insurance; Pauly opposes it) and epistemologically.6 Like Baker, Latsis and Repapis conclude that Arrow’s refusal to negate the complexity of the market for medical care as it operates in the real world came at the cost of weaker formalization, explaining the final victory of Pauly’s interpretation of moral hazard: “it is principally Pauly’s interpretation of Arrow’s paper that has dominated economics. The neoclassical tension that Arrow clearly struggled with when beginning the economic analysis of healthcare has been set aside.” (ibid, 14). In their view, it is because Arrow wants to describe institutions as they existed that his 1963 paper has not become a model for health economic analysis. This echoes Pauly’s own view on the controversy, as expressed in his foreword to the special issue of the Journal of Health Politics Policy and Law on Arrow, 1963: if moral hazard has something to do with trust and social institutions, then one needs a theory able to predict, rather than interpret, how trust and social institutions change over time (Pauly, 2001). Absent such a theory, Pauly believes that economists should rely on what they are able to do, which is to use prices and incomes to understand choices (and their subsequent impact on welfare).
13Our interpretation of the disagreement between Arrow and Pauly on moral hazard in health insurance differs from these historical accounts. We agree there is a deep philosophical divide between the two authors on the definition of sound economic analysis. Pauly followed closely Friedman’s (1953) influential epistemology which argues that the only orthodox tools of the economist are demand curves, prices, and incomes. Arrow, in contrast, warned in his 1963 paper that he wanted to include “the institutional organization and the observable mores … among the data ... in assessing the competitiveness of the medical-care market” (Arrow, 1963, 944). In the 1963 article, he repeatedly references sociological and ethical concepts, such as trust and social obligation. Remarkably enough, these references to sociological concepts are mostly made in reference to physicians and the production of medical treatments, not to the insured and their demand for medical treatments.
14We believe, however, that their differing views of the relative importance of prices and institutional mechanisms (trust, rationing) in the functioning of real-world markets, is not the direct cause of their different views on moral hazard. It is rather the other way around: it is because Arrow and Pauly differ on the nature of moral hazard in health insurance (and on the nature of the demand for health care) that they differ on the need to go beyond prices and quantities and include institutions in the analysis. To understand why, we need to revisit the treatment of moral hazard in health insurance in the two seminal papers: we will see that Arrow and Pauly differ on whether moral hazard is a market failure of health insurance; this difference of perception hinges on differences about the nature of coverage (what it is that health insurance should provide), which, in turn hinge on the nature of demand for health care.
15In Arrow (1963), moral hazard is a practical limitation of insurance that we would now describe as a market failure in the sense that individuals cannot get something (full coverage) they would be willing to pay for. Because of moral hazard, full coverage is not marketed; policies include co-insurance. It is a second best. Pauly (1968) can be read as rejecting the idea of market failure in health insurance due to moral hazard. Because coverage generates a welfare loss due to moral hazard and increases the cost of insurance, some individuals prefer not to buy coverage. There is no market failure because individuals “rationally” choose not to buy insurance given the premium offered. For that reason, we dispute the statement made by Baker (1996) and Latsis and Repapis (2015), that Pauly triumphed over Arrow, since many economists (not only health economists) would still characterize moral hazard as a source of market failure. We can therefore interpret Arrow as describing a market failure (even though he does not use the term) whereas Pauly states the market delivers exactly what individuals want.
16However, what matters most is the reason why they differ in their interpretation in terms of market failure: Pauly thinks that the product that the health insurance market should deliver is not full coverage but, rather, optimal coverage given the incentive to use more medical care when covered. In contrast, because people would purchase full coverage at the actuarially fair price but the market can’t deliver it, Arrow sees partial coverage as market failure. The difference between Arrow and Pauly is a matter of first-best counterfactual.
- 7 Pauly (1970) supports the idea that the ability of indemnity insurance (Variable Cost Insurance, su (...)
17This difference has to do with the different first-bests that the two authors have in mind: Arrow thinks that the first best is full coverage, which could be reached through contingent contracts (an indemnity for each combination of illness and severity). Because of informational constraints and transaction costs, however, insurers cannot offer these complex contingent contracts, making the first-best unattainable via the market. For Pauly, the idea that insurers could offer contingent contracts is a chimera; there is no and there cannot be such product. Contrary to property insurance, health insurance cannot be paid as indemnity but must be paid as a reduction of the price of medical care.7 We can characterize the difference of perception between Arrow and Pauly as follows: Arrow thinks that health insurance could work as any form of insurance were it not for problems of asymmetries of information on the health care market, whereas Pauly thinks that health insurance is intrinsically different and can only operate as price reduction.
- 8 A closed panel is a list of doctors the insured can visit to get re-imbursed. If the insured visits (...)
18This is why the deep difference between Pauly and Arrow is a disagreement on whether and why medical care is different from other goods and services: Arrow thinks that medical care is different because of radical asymmetries of information between physicians and patients (necessitating institutions to generate trust and not-only-for-profit behaviours on the part of suppliers); Pauly, on the other hand, thinks that while medical care is not different than standard economic commodities, it is different from the perspective of the insurer, because it is covered differently than most other financial risks. The following discussion by Arrow (1963) illustrates these differing perspectives on what health insurance supplies: Arrow describes three different modes of insurance payment, pre-payment (in kind), indemnity according to a fixed schedule, and insurance against costs (ibid, 962). Arrow notes that indemnity does not offer protection against financial uncertainty (of the cost of treatment), but he does not rule out pre-payment and, implicitly, suggests it is immune to moral hazard. Social institutions or mores prevent pre-payment to become popular because it is linked (or perceived to be linked) to closed panels of doctors8, which are rejected by patients. Pauly sees insurance against costs as the only viable method used in insurance, and one that generates moral hazard.
- 9 This sentence curiously refers to “normal” hazard, rather than moral hazard, even though it is quit (...)
19Therefore, the difference in admissible counterfactuals derives from a more fundamental difference in their views regarding demand for health care. Pauly’s embrace of demand-side cost-sharing to control utilization is consistent with a view of medical care as no different than standard commodities (at least to the patient; medical care is different to the insurer): price and quantity is all that is needed to understand utilization. Arrow, by considering pre-payment, brings physicians and allied professions (i.e., the supply-side) into the picture and shifts the onus of moral hazard to the supply-side of healthcare. In this view, the crucial moral hazard problem is physician agency, the way physicians deliver care and create treatments for their patients. As a result, Arrow views moral hazard on the demand side as a practical limitation that can be controlled by the physician: “To some extent the professional relationship between physician and patient limits the normal [sic]9 hazard in various forms of medical insurance.” Confirming the focus of Arrow (1963) on the supply-side, the special 2001 issue of the Journal of Health Politics, Policy and Law devoted to Arrow’s paper does not have a single article discussing demand-side moral hazard, the overuse of health care services by patients as a result of insurance. The only references to moral hazard of the insureds are in Pauly’s foreword (Pauly, 2001) and by Arrow himself in his response to comments (Arrow, 2001).
20Based on these different conceptions of medical care markets, we can interpret the differences on the role of institutions and mores. Why is utilization determined mostly by physician agency, which, simultaneously, is the main potential source of moral hazard and the best way to limit it? Because, according to Arrow, demand for care is not only derived (we want health, we buy health care), it is also delegated: it is the doctor who makes decisions on the types and quantities of care for a patient (here, Arrow anticipates the supplier-induced demand hypothesis). As a result, the patient’s sole decision is to choose a doctor. Demand-side moral hazard boils down to the lack of effort from the insured in selecting a cost-conscious physician; there is a moral or social aspect to this decision because this effort is shaped by a patient’s social network: they will choose a physician their friends and colleagues find acceptable. This is why Arrow (1968, 538), in his response to Pauly’s comment cites “the willingness … to behave in accordance to commonly accepted norms” as one of the three ways to control moral hazard. Of course, the other two ways have to do with providers: rationing schemes on the insurer part (utilization review) and professional ethics (the relations of trust between patients and physicians).
21Pauly rejects such a central role of the physician in the demand for health care, arguing that “Arrow appears to consider moral hazard as an imperfection, a defect in physician control, rather than a simple response to price reduction. He does not consider the direct relationship which exists between the existence of moral hazard and the validity of the welfare proposition” (Pauly, 1968, 535). This exchange, we believe, underscores that the main difference between Arrow and Pauly lies in a different conception of the nature of the demand for medical care: is it mediated by a professional who has the relevant information to make the decisions (Arrow), or does it reflect a “direct” link between a consumer and a product, as for any commodity (Pauly)?
22These differing views regarding the role of physician as mediator between patients and treatment is even reflected in their use of game theory. Whereas Arrow draws on tacit cooperation as his game of reference, Pauly refers to a version of non-cooperative games. Specifically, Arrow cites the notion of a focal point in Schelling’s theory of tacit games (Schelling, 1960) to support the idea that players in a game can collaborate (“through convergent expectations”) even though they cannot directly communicate, by using “clear and prominent signals [that] force patterns of behavior which are not in themselves logical necessities for optimality.” (Arrow, 1963, 966). Pauly, in contrast, uses the Prisoner’s dilemma to explain why all individuals have a rational interest in following the price incentives, even though it means higher premiums.
- 10 See also Joseph Newhouse in the introduction to his article testing the monopolistic (or not) behav (...)
23This difference in their views on demand is in turn the product of their differing views on the nature of medical care. Arrow (1963) is very explicit that, from a patient’s perspective, medical care is different: the product is the relationship and, due to the complexity and uncertainty of the effect of treatment on outcomes, patients cannot evaluate the quality of services delivered, even ex post. In this, Arrow follows the tradition of early health economics, that can be found, for instance, expressed in S. Mushkin (1958):10
Efficient organization of economic resources for health, guided by the consumer’s preferences, depends upon the consumer’s knowledge and the extent of his education. Despite persistent efforts to educate consumers they reveal considerable absence of accurate knowledge about the quantity and quality of health services required. The nature of the medical service itself, and its intangible character reinforce the consumer’s lack of knowledge about his purchases, and impede a rational choice that could guide the allocation of resources. (Mushkin, 1958, 787, emphasis added)
24It is because of this complexity and uncertainty that Arrow concludes medical care cannot be analyzed based on prices and quantities alone: “The logic and limitations of ideal competitive behavior under uncertainty force us to recognize the incomplete description of reality supplied by the impersonal price system” (Arrow, 1963, 967). For Arrow, this has policy implications: it provides a rationale for governments to buttress the peculiar institutions shielding medical and allied professions from competition.
- 11 Pauly uses a standard argument in the public choice literature, developed by his mentor, J. Buchana (...)
25To the contrary, Pauly (1978), rejects the idea that medical care is different and that the peculiar institutions of the market for medical care can be justified by the failure of the demand curve to represent the value to patients. He suggests that causality runs in the other direction: these professional institutions that Arrow thinks are necessary to create a relation of trust when information is asymmetric are in fact the source of the meaninglessness of the demand curve, rather than their consequence. Asymmetric information is used as an excuse for monopolistic institutions rather than these institutions being a rational response of society to a true market failure.11 One such institution that, according to Pauly, “distort[s] demand curves” (Pauly, 1978, 19) is group health insurance, favoured by tax exemption: on the individual insurance market, insureds are aware of the trade-off between coverage and premiums and can trade a higher co-insurance for a lower premium. In group health insurance, due to heterogeneity of preferences and choices being set by the median voter (represented by the employer as sponsor), there is no such link, and demand curves are distorted. Pauly argues that moral hazard due to over-insurance is the main source of welfare loss on the market for medical care. In Pauly (1986), he describes the issue as a regulatory failure: without the tax exemption favouring group health insurance there would be much less moral hazard and much less welfare loss. He thinks that parametric incentives such as co-insurance rates to control demand, and more, rather than less competition among providers are to be preferred (Pauly, 1970).
26Arrow and Pauly’s positions and definitions of moral hazard in health insurance can be summarized as follows: for Arrow, moral hazard is a practical limitation of the market for health insurance that can be overcome through social institutions (it is a social/moral issue): the practical limitation is that contingent contracts cannot be written and insurance must therefore pay off through a reduction in the cost of health care; it is a minor limitation, though, because providers make most of the decisions on behalf of their patients and providers can be regulated and organized in a way that guarantees they control their patients’ level of utilization (social institutions and norms create a trust relationship between physicians and their patients). Therefore, more institutions to support provider control of utilization and limit moral hazard, and to protect providers against competition, are welcome. For Pauly, moral hazard is a natural law of economics and explains why individuals (who anticipate the effect of moral hazard on demand) choose partial (or no) coverage when insurers cannot write perfect contingent contracts (due to the nature of health insurance). Because, for him, moral hazard reflects the natural functioning of a market, the only way to “address” it is by letting the market offer baskets (premium and coverage rate) among which consumers can pick what is optimal based on their preferences. As a result, Pauly sees partial coverage as the first-best solution to moral hazard, whereas Arrow sees partial coverage as a market failure due to transaction costs and asymmetrical information that can be solved by compulsory insurance and robust institutions to control moral hazard. Importantly, their two views—“practical limitation” vs. “natural law of economics”—hinge on diverging vision of the nature of demand for health care: if patients’ choices are limited to selecting a doctor who then makes treatment decisions on their behalf, then moral hazard is a moral issue (Arrow); if consumers are sovereign then moral hazard is natural and partial coverage is rational (Pauly).
27Much of the debate on moral hazard after 1968 revolved around this question of whether medical care is different. “Mainstream” health economists who followed Pauly’s views sought empirical estimates of the reactivity of health care utilization to changes in cost-sharing, so as to evaluate the welfare loss of health insurance due to moral hazard (Pauly, 1969; Feldstein, 1973; Manning et al., 1987; Feldman and Dowd, 1991; Manning and Marquis, 1996). Health economists whose views aligned more closely with Arrow (whom Evans (1976), called the “broad” economists, as opposed to the “narrow” in the mainstream) doubted the sovereignty of consumers of health care services, mostly based on their inability to assess the quality of care ex post. Of course, even mainstream economists were ready to concede that health care was not fully standard, and patients were not fully informed. However, they thought that patients could be provided with more information on health care services, and thereby bring health care markets closer to standard ones. They described medical care as an experience or reputation good rather than what is now known as a credence good that requires, as Arrow described it in 1963, a relation of trust between the supplier and the customer.
- 12 This result has been disputed, and still is. A key concern, even at the time of the experiment, is (...)
28It is broadly accepted among health economists that patients react to the price of health care (elasticity is not zero), and the RAND Health Insurance Experiment (HIE) found that the effect of co-insurance (leaving a percentage of the cost of the service uncovered, to be paid by the patient) was mostly on the number of visits or stays, not so much on the portion of health care use that is truly delegated to the physician (intensity of treatment).12 That elasticity of demand for health care is not zero, however, is not the most important point: what matters is the interpretation of the changes in utilization (including initiation of treatments) when co-insurance varies. Even before the results of the HIE were known, the “broads” raised doubts regarding the normative interpretation of the demand curve as reflecting the surplus of a sovereign consumer: the value of health care services is not in the price patients are willing to pay for it, but in the amount of health it produces for patients (or in the appropriateness of the service from a clinical perspective), opening the way to the extra-welfarist school of thought in health economics. The extra-welfarist school drew empirical support from what is now known as the “offset effect” (with results becoming available in the 1990s). As a response, the “narrows” developed the concept of “behavioral” hazard, where some of the response of patients to changes in prices is not rational, but the demand curve remains a sufficient statistic for welfare.
29A strong proponent of the vision that medical care is different is the supplier-induced-demand hypothesis, first expressed by Evans (1974). Supplier-induced demand (SID) holds not only that physicians influence patient demand (a view shared by nearly all health economists, e.g., Feldman and Morrissey, 1990) but also, more subtly, that physicians can systematically shift patient demand in response to changes in the economic environment physicians face, such as those caused by changes in fees or in the supply of physicians (Evans, 1974). Central to this is the view that patients cannot evaluate the quality of care, even ex post, in the sense that they cannot infer back from outcomes to quality. As a result, there is no such thing as an autonomous demand for health care, and interpreting the area below the demand curve as consumer’s surplus is problematic if the SID hypothesis holds true.
- 13 Pauly and Satterthwaite (1981) argue that limiting the number of providers, through licensure and r (...)
30Pauly (1978, 1988) disputed the SID hypothesis: he argues that, even though patients are not and will never be medical experts, meaning they cannot assess the medical appropriateness of a specific service, they are able, in some instances, to infer a link back from an outcome that is meaningful to them and the quality (effort and skill) of the provider. At least, patients can make such an inference for a medical service they use on a regular basis (or that their friends use on a regular basis): pediatric care, natural delivery, care for chronic conditions, prescription drugs, routine dental care, well-care. For Pauly, these medical goods are either experience goods or reputation goods, for which the demand curve works exactly the same as for any experience good.13 Pauly (1988) further argued that four developments would make the market for medical care more competitive and less different, giving more strength to the standard normative interpretation of the demand curve for health care. Among these developments, two are related to information on quality of care and its commodification: the development of Health Maintenance Organizations, whose closed panels and group practice lead to an increase in available information on both the providers themselves and care outcomes more generally, and the emergence of a market for information on outcomes and processes that could lead medical care toward the perfect competition model. Pauly acknowledged there are difficulties associated with the production, diffusion, and pricing of information on quality of care (preferences are heterogeneous, 3rd party payers might not want to disclose that lower cost means lower quality, and too much emphasis on cost might be harmful if patients seek too much low-quality services), but he still posits that the share of medical goods that would become either experience or search goods will increase in the future, rendering the peculiar institutions advocated by Arrow unnecessary.
- 14 Feldman and Dowd (1993) summarize it nicely: “We believe that measurement of consumer welfare shoul (...)
31The debate around the SID hypothesis and its empirical validation is still alive and relevant to the field of research on physician agency but the SID hypothesis cannot explain why individuals change their decision to initiate a treatment episode when the co-insurance rate they face change (as was established by the RAND HIE, Manning et al., 1987). More detailed results of the RAND HIE prompted the second debate around the nature of demand for medical care, centred on the interpretation of the demand curve. Culyer and others were developing an extra-welfarist approach to health economics in which the value of medical care services was not based on revealed preferences and willingness-to-pay (the price paid by the consumer) but on their effects on health (e.g., Culyer, 1971; Torrance et al., 1972). Evans (1983, 1984) was also calling into question application of standard welfare economics methods to health care, arguing in part that care utilized and care needed were not necessarily the same thing (nor was care forgone and care not needed): “What does not follow from observed price sensitivity by individual patients, however, is … that any changes in utilization which occur will be among the least needed forms of care” (Evans, 1984, 90). The demand curve did not provide a reliable guide to the welfare effects of demand-side policies such as cost-sharing that affected utilization. Evans argued that third-party assessments, based on high-quality clinical evidence regarding the effectiveness of health care services, could inform the assessment of the “welfare” effects of change in utilization. Pauly (1978) rejected the idea that necessity or appropriateness can be assessed independently of individual preferences, as was commonly done in the literature on small-area variations in rates of care. He suggested an alternative definition, inspired by the logic of cost-benefit analysis: benefit is a weighted sum of gains in longevity, averted illnesses, and peace of mind or reduced pain. In his view, non-health outcomes are perfectly legitimate, and the weights given to different types of outcomes must reflect the preferences of a fully-informed patient, by which he meant a patient with as much information as her physician (otherwise, the patient can be influenced and her demand does not reflect her willingness-to-pay for the services she receives).14 But extra-welfarists go further and state that a perfectly informed patient should always select services on the basis of their effectiveness and, as a result, there should not be any difference between tastes and health outcomes. They acknowledge that there might be a conceptual difference between allocative efficiency (what consumers want) and effectiveness (what they need) but claim that the only difference between the two is information, the fact that patients are less informed than medical experts (those who evaluated effectiveness). Contrary to Pauly (1979), who states that “peace of mind” is a valid reason to seek care, they assert that a perfectly informed patient would never knowingly use ineffective, unnecessary care (Rice, 1992).
- 15 We single out these two papers but it must be kept in mind that they are the product of a vast effo (...)
32This extra-welfarist line of argument received empirical support from some RAND HIE results and the broader literature on the “offset effect”. Two empirical studies based on the RAND HIE (Lohr et al., 1988, for ambulatory care, and Siu et al., 1986, for hospitalizations)15 demonstrated that individuals facing higher co-insurance rates had a lower probability to start an episode of care both for diagnoses for which care was deemed (by medical experts) to be highly effective and for diagnoses for which it was deemed not effective. These and similar results led Thomas Rice to dispute the rationality of the demand for health care (Rice, 1992). For Rice, such evidence demonstrates that individuals do not reduce their demand for services of “marginal value” when their co-insurance rate increases, where Rice takes “value” as meaning “objective effectivity”, not “subjective value, whatever the true—medical—benefit” of the service.
33Feldman and Dowd (1993, 194) call Rice’s criticism “serious” because it calls “into question one of the profession’s widely-accepted and utilized tools”. Their most fundamental objection to his new framework is on the definition of “ineffective services” as “diagnoses for which care is not more effective than self-care”. They argue that patients are right to seek care for diagnoses categorized as non-effective because patients consult mostly to get information on the diagnosis, which they don’t know in advance and need a professional to determine what they suffer from. They cannot tell in advance whether care will be effective or not and it might be perfectly rational to start such an episode of care.
34The late 1990s saw renewed interest in demand-side cost sharing, particularly in the form of high-deductible health plans (HDHP) with 100% co-insurance up to a spending limit, followed by low co-insurance coverage (possibly with a maximum dollar expenditure above which co-insurance fell to 0). Interestingly, the new regulation enacted in the early 2000s for HDHP follows approximately the plan recommended by welfare analyses of the results of the HIE, with one exception: preventive services were excluded from the co-insurance and the deductible. Newhouse (2006) argues that ambulatory care and many drugs for chronic diseases should be exempted as well (note that these services were considered reasonable experience goods by Pauly in 1978).
- 16 To which can be added several publications after Newhouse (2006): two systematic reviews, Freeman e (...)
- 17 It might also be the result of an over-cautious and quantitative way to present the findings of the (...)
35The reason for recommending the exemption was new empirical evidence on the “offset effect” of low compliance: when patients do not comply with their chronic diseases treatments because of cost-sharing, they cost more in the future as they experience more adverse events such as hospitalizations or their health deteriorates, possibly even resulting in death (Soumerai et al., 1991, 1994; Tamblyn et al., 2001, and Hsu et al., 2006.)16 These findings, which confirmed the views of the broads on how demand for health care works, were not accepted immediately by mainstream health economists (the “narrows”) because they contradicted a widely circulated result of the RAND HIE according to which co-insurance had little impact on general health for the general population. However, given the high-quality of the studies and their consistent results, the result of the RAND was now seen as an outlier (possibly due to its healthy, working-age study population, systematic attrition, and its relatively short-term follow-up).17
- 18 Pauly and Blavin (2008) is another attempt at salvaging the mainstream model, but without behaviour (...)
36This empirical result could have vindicated key elements of the Arrow-Evans-Rice viewpoint on the nature of medical care: faced with higher co-insurance, patients cut effective as well as ineffective services, and this translates in poorer health. In response, however, the mainstream school developed a new conceptual framework to account for the offset effect: inconsistent time preferences could explain the lack of compliance of patients when faced with high levels of cost-sharing. Individuals are still rational in the long-run and their demand for health care should be taken seriously, but behavioural anomalies prevent them from doing it in the short run: typically, they will not comply with drug treatments to control their chronic condition if it does not have immediate effects. Newhouse (2006) describes this behavioural ad-hoc addition as an “elaboration on traditional theory” rather than a challenge to the mainstream, welfarist model of the rational health care consumer. It can therefore be described as a way to salvage the mainstream model of moral hazard in the face of new evidence of offset effects. Newhouse traces the origin of his new framework to the concept of benign moral hazard, developed in 1990 by Pauly and Held (1990).18 A Google Scholar search shows that Pauly and Held (1990) remained quite obscure for a long time before being cited abundantly in 2005-2009. Benign moral hazard exists when more elastic demand requires lower co-insurance, and these situations all have to do with the offset effect and the intrinsic, medical value (here translated into dollars, hence consistent with a CBA) of the treatment. Newhouse (2006) starts from this concept, devised for value-based health insurance—covering more generously services that have more medical value (Chernew et al., 2007)—and enriches it with a behavioural component.
- 19 This is the meaning of sufficient statistic: in standard moral hazard, the patient uses the service (...)
37“Behavioural hazard” builds on Newhouse’s intuition. The term was suggested in 2012 in a working paper by Baicker et al. (published in 2015). The idea is that in the “standard” (or mainstream) moral hazard model, the demand curve is taken as a “sufficient statistic” for welfare (Baicker et al., 2015, 1): “When looking at a change in copays, we can draw welfare conclusions without ever measuring the health impacts because we infer such impacts: if people optimize perfectly, health benefits equal copays at the margin.” However, if individuals make mistakes when they decide what and how much services to consume, the inference is not valid. Authorizing consumers to “make mistakes” (including due to wrong beliefs on the effectiveness of medical care) looks similar to the way Evans and Rice would have described their alternative framework for evaluating welfare in health insurance (Evans or Rice are not cited in Baicker et al., 2015), but it is in fact subtly different. Whereas for Evans and Rice poor patient decision-making, rooted in a lack of information, rendered the demand curve for health care normatively meaningless, the behavioural approach maintains the idea of an underlying well-informed (in the long-run) rational demand for health care, but adds to it a perturbation reflecting lack of self-control on the part of the consumer. According to this framework, individuals know what they should do, but fail to act on this knowledge due to emotions or false perceptions or, as in Newhouse (2006), time inconsistencies. It is, admittedly, difficult to see how this differs from consumers making poorly-informed choices in the Evans-Rice tradition, however a major difference is that it allows the econometrician to specify demand for health care as a function with two components: the rational, well-informed, meaningful (sufficient statistic) demand of the traditional model, and a random perturbation the parameters of which can be retrieved from observation.19 Clearly, this is an effort to salvage the foundations of the mainstream approach from damages inflicted by empirical evidence on the offset effect. It finds a new, smaller, space for the “sufficient statistic”, based on a re-interpretation of the empirical data and some econometric sleight-of-hand.
38This section has shown that the initial controversy on the moral or natural nature of moral hazard hinges on conceptions of the nature of demand for medical care: is it supplier-influenced demand by a poorly informed consumer or autonomous demand of a sovereign consumer? Logically, health economics investigated the nature of demand for care from a theoretical and empirical perspective, focusing on the critical issue of asymmetric information between patients and physicians. Few health economists dispute that patients are not fully informed, and less informed than their physicians, but the question is the extent of their dependence on physicians and their ability to assess quality of value of care: can we assume that demand can be interpreted as reflecting the value of services consumed or should it be established first that changes in utilization often have little or no impact on health outcomes?
39The empirical result emerging in the 1990s and 2000s that higher co-insurance deteriorates health (the offset effect) and contradicting the conventional presentation of previous empirical results from the HIE (that co-insurance had no effect on health) prompted mainstream economists to reconsider their assumption of a perfectly rational patient and to develop a new behavioural framework in which patients are well-informed but do not always use the information they have properly. The former opposition between a “moral” moral hazard and a rational one has now become one between a moral and a behavioural conception of moral hazard. In the former, social norms can affect the price elasticity of demand for episodes of care (and suppliers can control intensity of episodes), whereas in the latter, moral hazard is a rational response to changes in price, with some behavioural perturbations.
40A more recent area of disagreement revolves around the nature of the demand for health insurance. Chronologically, it erupted at the end of the 1990s with the two publications in the Journal of Health Economics by John Nyman on the value of health insurance, followed by a 2001 exchange in the same journal between Nyman on one side and Å. Blomqvist, W. Manning and S. Marquis on the other side, and then by Nyman’s publication of a full-length book in 2003. The controversy was even declared settled in 2010 by Mark Pauly, who publicly stated that he and Nyman agreed (Kelman and Woodward, 2013), though as Kelman and Woodward remark, it is not entirely clear why Pauly made such a statement since the views of these two economists still differ markedly. Pauly reiterated an often-made comment that Nyman’s New theory was true only in some marginal cases and that, in those marginal cases, it was not truly original, but only a special case of the conventional theory. We do not agree with such a characterization, and we think that Nyman’s and Pauly’s views on moral hazard in health insurance differ strongly because they are based on diametrically opposed views on the motives for the purchase of health insurance.
41In the introduction to his book on the theory of demand for health insurance, Nyman tells why he became interested in the topic. Following the debates on the so-called Clinton plan to implement universal health insurance in the US, Nyman read numerous stories of non-insured or poorly insured Americans who could not afford care that they obviously needed and wanted, and paid for that lack of access with pain, disability, or sometimes the loss of a family member. He then compared those testimonies to the conventional welfare analysis of moral hazard in health insurance, which considers the level of health care utilization of the uninsured as the optimum and any level of utilization beyond that as producing a deadweight loss. He concluded that economists were missing something important.
42According to the conventional model of insurance (including, and, in fact starting with, Arrow (1963), himself building on Friedman and Savage (1948)) the sole motivation for purchasing health insurance is risk avoidance—in the case of health care, the financial risk associated with the cost of a treatment when sick. Under marginally decreasing utility of money (risk aversion), the utility of expected income (income minus an actuarially fair premium) is greater than the expected utility of these two levels of income with and without the loss. As a result, if offered insurance priced at the actuarially fair premium, the best option for a risk-averse individual is to purchase full coverage. The insurer can even charge more than an actuarially fair premium up to the individual’s risk premium, at which point the individual would be indifferent to buying insurance. On the risk-avoidance side of this model, medical insurance can be described as contingent-claims insurance (the cost of treatment is fixed and known in advance) and individuals benefit from certainty in income in both states of the world (rather than a lottery where they can lose income if they get sick). Moral hazard is an issue in the insurance market if the cost of treatment once covered is greater than the cost of treatment when uninsured, and, implicitly, the cost of treatment when uninsured is taken as the “true” level of treatment (the indemnity that the insurer should transfer to the insured when sick). As described in section 2, this is where Arrow and Pauly part and we now detail the Pauly interpretation of this effect of coverage on utilization (the conventional interpretation): Conventional welfare analysis consisted in estimating a trade-off between the benefit of risk avoidance (valued as the risk premium) and the welfare loss of moral hazard. Balancing this trade-off leads to an optimal rate of cost-sharing given the elasticity of demand for health care. Because all individuals are the same in this model, including having the same level of income, income falls out of the picture, even though it is well known that some treatment costs may exceed total wealth for a large portion of the population.
43Nyman’s objection to this story is that, precisely, the share of one’s budget going to health care spending should be part of the discussion on the demand for coverage: as the testimonies during the discussions of the Clinton reforms showed, individuals wanted to spend more on health when sick, but could not, because their level of income did not allow them to do so. The level of spending of at least some among the uninsured was certainly not optimal and they would have been happier if they could have transferred income to themselves when sick and used that extra income to spend on health care.
- 20 If U is the Bernoulli utility function, Y income, T cost of treatment and p the probability of bein (...)
44Nyman contends that the rationale for purchasing health insurance has nothing to do with risk-aversion, and anything to do with the diminishing marginal utility of income. This looks paradoxical if, and only if, we assimilate the two, but Nyman claims that the latter can be true (and is likely true, as it is very intuitive) without implying the former. For instance, individuals might purchase insurance and gamble at the same time, which would not make sense if the purchase of insurance could be explained by risk-aversion. Nyman (2003) shows that the difference between the risk-avoidance motivation and the motivation based on diminishing marginal utility of income is a matter of interpretation of movements along the Bernoulli utility function (Nyman, 2003, 128-133): in the risk-avoidance story, individuals prefer a fixed income (paying the premium) to an uncertain one (full income versus income minus cost of treatment if sick). In the diminishing marginal utility story, individuals are ready to pay (a premium) to play a lottery that will pay them the cost of treatment if sick. In both instances (insured or not) of the diminishing marginal utility story, individuals play a lottery (hence, they do not avoid risk), but the “insurance lottery” can be preferred because, provided the cost of treatment is large relative to income, the gain in utility of the payback from insurance is seen from the perspective of a much lower income, where marginal utility is much greater.20 In this story, “the medical spending is really not part of the insurance contract” (Nyman, 2003, 133), but “the defining characteristic of insurance is that the contract is set in a context where the event that triggers the payoff also moves the consumer to a lower point on her Bernoulli utility function”. With such a model, risk-lovers can still buy insurance, if the cost of the treatment (damage) is large enough that the transfer of income when sick (unlucky) improves utility of income enough. It is worth noting that this aspect of Nyman’s theory has been little discussed or even noticed (an exception being Eisenhauer, 2004).
45The trade-off is not between the risk premium and the loss due to increased consumption of care once covered, as in the conventional analysis, but between the beneficial effect of the income transfer and the portion of moral hazard that is linked to using a price reduction to transfer income. The latter can be seen as a transaction cost to benefit from the income transfer. If an individual purchasing insurance increases the co-insurance rate, they decrease the income transfer when sick and in need of income. On the other hand, if they reduce the co-insurance rate too much, a pure price effect will lead them to spend too much on health care relative to other goods, within their budget, and this will cost them in well-being. The latter is a conventional Pauly normative interpretation of moral hazard but it applies to a portion of increased health care utilization only, the other portion of that increased utilization being welfare improving and, as a result, on the other side of the trade-off. The optimal co-insurance rate, in Nyman’s theory, maximizes the welfare gain of the income transfer net of the welfare loss of moral hazard due to substitution.
46This income effect leads to questions regarding the demand curve on which to base the welfare analysis of moral hazard: Pauly (1968) uses a Marshallian demand curve for which the entire change in utilization reflects a welfare-reducing price effect. More recent approaches (Manning and Marquis, 1998) use the Hicksian demand curve, where the change in utilization is decomposed between an income and a price effect, the latter only being welfare detrimental. Nyman suggests a new decomposition of the income and price effects, in which the income effect is larger than in the Hicksian model. The appendix summarizes the differences between the various decompositions, which revolve around the definition of compensated income (the income that would make the individual as rich after as before the change in the price of one good).
47Nyman’s theory of demand for health insurance, which holds that individuals purchase insurance to protect themselves from income losses rather than from uncertainty per se, has strong normative implications for our understanding of moral hazard. Since individuals purchase an income transfer when they buy insurance, the portion of moral hazard that is due to the income effect is welfare beneficial: this is precisely the benefit the consumer of insurance wanted (a transfer of income from the healthy to the sick state). The trade-off for the individual consumer is now between this income transfer and the increase in utilization that is associated with it, and the transaction cost of transferring income through a reduction of the price of health care. Last, the New theory also predicts that this welfare loss of moral hazard associated with the change in utilization due to the price effect is lower than in the conventional (Hicksian) decomposition.
- 21 The generally muted reception to the New theory resulted at least partially from the way it was int (...)
- 22 Zeckhauser (1970, 10, note 2): “It is not only because individuals’ utility functions for money dis (...)
48The reaction among many health economists and welfare economists to Nyman’s New theory was to dispute its novelty and its empirical relevance.21 It is true, in fact, that income effects have been acknowledged by economists from the start of the moral hazard discussion. For instance, Zeckhauser (1970) made several references to affordability and the value of being richer when sick.22
49Overall, it was conceded that income transfer could be a motive in some very specific cases, when the cost of treatment is greater than total income, something known as the “access motive” (Nyman, 1999b), which was treated as interesting but of no real empirical or practical relevance for most health care consumptions. In reacting to De Meza (1982), which highlighted insurance as a way to transfer income from the healthy to the sick state (and a better one than saving or borrowing), Pauly (1983) did not reject the income effect, but rather downplayed its empirical significance. He argued that a positive income effect for individuals with severe illnesses would be compensated (perhaps even more than offset) by a negative one due to the fact that individuals with minor illnesses or almost no health issues are made poorer by the payment of their health insurance premium. He noted, nonetheless, that a theory of demand for coverage against such risks is lacking: “It is nevertheless true that the relevant theory, empirical evidence, and policy analysis for moral hazard in the case of serious illness has not been developed. This is one of the most serious omissions in the current literature.” (83).
- 23 In their response to Nyman, Manning and Marquis (2001) claim that he stated they did not use a comp (...)
- 24 Another empirical puzzle in Manning and Marquis (1996), identified by Nyman (2003, 162, note 2) is (...)
50The empirical literature also minimized the practical importance of accounting for income effects in the decomposition of moral hazard: Manning and Marquis (2001) and Blomqvist (2001) concede that income effects exist, but argue that they do not change the main conclusion on the welfare effect of moral hazard. Manning and Marquis (1996) use a Hicksian decomposition23 and an estimate of 0.22 for the income elasticity of health care utilization derived from the RAND HIE to show that most of the cost-sharing-induced change in health care utilization in the RAND was due to a substitution effect. Nyman objects to their decomposition for two reasons: first, the income elasticity is derived from the whole population, and not the sick only, and Nyman and Maude-Griffin (2001), using an income elasticity of 0.4 (from Feenberg and Skinner, 1994, based on aggregate level changes in GDP and health care use) estimate the welfare cost of moral hazard to be much smaller than Manning and Marquis.24
51Nyman (1999a) raises another issue with estimates of the income elasticity of the demand for health care: he claims that the true income elasticity cannot be derived from revealed preferences (health care consumption at various levels of income, ceteris paribus), but only the change in utilization that occurs when individuals receive a large income transfer when sick, including one that would allow them to spend more than their level of income:
The ideal social experiment to determine the price and income transfer effects of private insurance would be to identify voluntary purchasers of a price-payoff contract (say, a 0% coinsurance plan with an actuarially fair premium), to develop a contingent claims plan with contingent payoffs equal to the various expenditures under the 0% plan, and then to randomize purchasers among the two plans and no insurance. The observed difference in health care spending between the two plans is the pure price effect of insurance, and the difference between spending with the contingent claim plan and spending with no insurance is the income transfer effect (Nyman, 1999a, 298).
52The RAND HIE analyzed an exogenous price change because, historically, it was meant to simulate a national price-reducing health insurance plan in which individuals would benefit from a reduction in prices paid for by others (taxpayers). However, when individuals make actual choices (on the private market for health insurance), the income transfer is endogenous and an integral part of the quid-pro-quo of the value of insurance, including large income transfers for catastrophic illnesses and access to treatments that extend life but are not affordable given an individual’s available resources (the access motive, Nyman, 1999b). In their response, Manning and Marquis (2001), admit that their 1996 estimation excluded the access motive. However, they reiterate that the only way to estimate an income elasticity is through revealed preferences and they even claim it is “the economist’s ceteris paribus” way of estimating an income elasticity.
53Even though ideal experiments are hard to implement, quasi-experiments of large income transfers (unconditional on behaviours or consumption) can provide insights on the income effects at play in health care utilization. Such quasi-experimental designs have been used in fields other than health (Autor and Duggan, 2007; Chetty, 2008; Costa, 1995) and find income elasticities much larger than the 0.22 value used by Manning and Marquis (1996). Overall, the income elasticity of demand for health care is certainly complex to estimate, which may be a reason why it has been downplayed in the economic literature, both in health and in labour supply (Autor and Duggan, 2007). It might also be linked to the fact that it re-introduces distributional issues in the welfare analysis, something the canonical welfare analysis of economists avoids (Arrow, 1963).
54Nyman’s theory is a departure from the conventional welfare analysis of moral hazard, but neither does it align with Arrow’s analysis of moral hazard. Nyman shares with Pauly a belief that patients are generally informed although his views often assume “objective”, external measures of the value of health care rather than subjective ones revealed by “willingness-to-pay”. Nyman is quite clear in his 2003 book that his theory does not align with Rice’s critique of Pauly: “the theory described in this book is diametrically opposed to Rice’s theory” (Nyman, 2003, 151). Indeed, Nyman’s consumer is even more rational than Pauly’s, sharing the level of information on health care and health that Zeckhauser (1970), or Blomqvist (2001), assume for their consumers of health insurance: not only do they know exactly what the value of health care is, but also they know it because they are aware of the contribution of health care to the production of health. Consumers seek a transfer of income through insurance to buy health care that they will then transform into health. The transfer enables them to produce health when they need it. However, the third-party analyst cannot infer that value of health care from what individuals actually consume (especially the uninsured) precisely because they are constrained by their budget. This is especially the case for catastrophic illnesses requiring costly treatment: an uninsured may well forego treatment, or buy only partial treatment, because the total cost of the full treatment would exceed their total wealth by orders of magnitude. But this does not mean that what they spend represents its value for them. Consequently, Nyman must rely on “objective” measures of the value of health care and treatments, based on Quality Adjusted Life Years, to estimate the access motive of insurance or the value of the income transfer when income elasticities are not reliable. Similarly, it is crucial for Nyman’s theory that co-insurance has an effect on health and he cites numerous studies documenting such effects (though they are of variable quality).
55Overall, therefore, positions on moral hazard in health insurance can be mapped along two axes of opposition: the first axis opposes welfarists, who believe that normative assessment of utilization should be based on the conventional welfare economic framework and that patients are well informed on the quality of health care services (which are experience or reputation goods) versus extra-welfarists, who reject the idea that the value of health care is reflected in willingness-to-pay. This latter camp aligns with the institutionalist tradition elucidated by Arrow in 1963. In contrast, the welfarists divide between those who view risk avoidance as the main motive for purchasing insurance and those who follow Nyman’s analysis and view the income transfer as the main motive for purchasing insurance. Both of these welfarist groups believe that willingness-to-pay is the same as the value of health care used, but use different demand curves (this is the second axis of opposition): the conventional theory sees the income transfer of insurance as exogenous whereas Nyman’s theory sees it as endogenous and claims that the marginal utility of income is state-dependent. Hence three schools of thought populate these two axes: the conventional mainstream approach sees moral hazard as a characteristic of the good “health insurance”, which partially explains why rational consumers do not buy health insurance; the Nyman approach sees moral hazard as comprising a beneficial income transfer and a deadweight loss due to the fact that the income transfer is provided through a reduction in the price of health care (this is a transaction cost: it is impossible to transfer income in health insurance without generating some substitution effect because pure contingent-claims contract are not feasible). The third school, which follows Arrow’s institutionalist approach, shares with Nyman’s approach the idea that transaction costs are responsible for the lack of coverage but sees moral hazard as largely independent from these transaction costs and as a minor practical limitation.
56This history of how moral hazard in health insurance is conceptualized and understood shows that, if health economists agree that health care utilization responds to cost-sharing, there is no consensus on the normative interpretation of this empirical fact.
57Overall, there are three main positions on moral hazard in health insurance: the mainstream (with some recent behavioural adjustments for beliefs and information), the institutionalist (extra-welfarist economists who think that medical care is different and that the value of medical care derives from its effect on health), and what we will call the “state-dependent utility” (SDU), based on the idea that individuals purchase insurance to transfer income to the state where the marginal utility of health care is large.
- 25 In the case of Nyman’s new theory, it is fair to say that it does not really involve more interpret (...)
58The institutionalist and the “state-dependent utility” approaches are not compatible, since the former sees patients as ill-informed and unable to assess the value of health care whereas the latter thinks that patients can assess the value of health care services (measured by the marginal effect of health care on health, and the marginal utility of health) and make rational decisions to buy coverage on the basis of such assessment. However, institutionalists and SDU share a common epistemological position differing from the mainstream approach: whereas mainstream economics is parsimonious and highly predictive (all it needs are robust estimates of the causal effect of changes in the price of health care on utilization), both the institutionalist and “state-dependent utility” approaches rely at least in part on interpretation25 and story-telling (Feeny, 1988), and as a result, are less abstract methodologically and more “rhetorical” (McCloskey, 1983). Pauly objects to Arrow’s institutionalist stance (Pauly, 2001) because there is no reliable theory or model of the genesis of institutions. He adds that if institutionalists want to be taken seriously in health economics they should account for the dramatic changes in institutions within which health insurance and health care provision have been operating since 1963.
59Extra-welfarist institutionalists and state-dependent utility theorists tell very different stories, but their stories both imply a similar characterization, and empirical measure, of the value of health care services: the impact on health. In contrast, the mainstream welfarist approach is both parsimonious and circular (circular because parsimonious, and parsimonious because circular): rather than trying to define and measure value, it uses actual consumption behaviours to “reveal” preferences. Evans (1984) characterizes the definition of “necessary services” in Pauly (1979) as circular: “[T]o define patients responding to professional advice as fully informed, so as to give their utilization normative significance—whatever the observed level of use, it must be right—is a dodge worthy only of Pangloss.” (Evans, 1984, 84). Interestingly, such an objection of circularity mirrors the objection raised by Pauly against Arrow’s description of the emergence of social institutions in response to the specificities of medical care. In his JHPPL contribution, Evans (1997) attributes the circularity to a desire among mainstream economists to remain value-neutral, guided by the (objectively measured) revealed preferences of a representative agent motivated by prices only, but argues that: “Theory by itself does not, and logically cannot, provide a normative basis for policy prescriptions. … Normative judgments, in or out of economics cannot be derived from positive propositions alone, or in Archibald’s paraphrase of Hume: ‘No ethics in, no ethics out.’” (Evans, 1997, 455).
60The story that the institutionalist approach must tell is that of the value of health care services from an extra-welfarist perspective. In that sense, positions about the interpretation of moral hazard in health insurance are really positions about the way to measure the benefit of health care services. The economic history of how the profession embarked on measuring health benefits and developed the Quality-Adjusted Life Years is told by E. Forget (Forget, 2004) as an effort to create tools specific to health economics (cost-utility analysis as opposed to cost-benefit analysis, as used in the rest of economics). The story that the SDU tell is how marginal utility of consumption changes when individuals move from health to sickness. To do this, they have to compare utility across individuals or states, which leads to even more ostracism than an institutionalist stance, even though empirical work is now emerging on how to measure health-dependent utility at the individual level using self-reported happiness (Finkelstein et al., 2013).
This study benefited from research assistance from Valerie Ulep (McMaster). We also thank the two anonymous referees for their very helpful comments on a previous version of the manuscript.