Reconsidering the moral hazard-risk avoidance tradeoff

Journal of Health Economics25(2006)1005–1014Reconsidering the moral hazard-riskavoidance tradeoffଝJoseph P.NewhouseHarvard University,180Longwood Avenue,Boston,MA02115,United StatesReceived14June2006;accepted14June2006Available online24July2006AbstractThe received model for optimal demand-side cost sharing trades off moral hazard and risk avoidance. This model appears to lie behind recent increases in initial cost sharing,such as those embodied in Health Savings Accounts and Health Reimbursement Accounts.At the same time there is evidence that lower cost sharing for certain drugs can reduce future total health care costs and/or improve future health,and this may be true of other medical services as well.To the degree that individuals remain in the same common insurance pool,lower cost sharing that induces increases in certain services that reduce total costs,including future costs,represents a classic case for a subsidy and will minimize an employer’s labor costs.Even if total costs increase,the value of a change in health could increase more.In that case such a subsidy is consistent with recent work in behavioral economics for those with self-control problems.©2006Elsevier B.V.All rights reserved.Keywords:Moral hazard;Risk avoidance;Cost sharing;Optimal insurance contracts;Hyperbolic discounting Our traditional view of demand prices or cost sharing comes from several classic papers, among them(Arrow,1963,1968;Pauly,1968;Zeckhauser,1970).These papers frame demand prices as a tradeoff between moral hazard and risk avoidance,and that is how health economics textbooks present the problem,for example,(Phelps,2003),chapter10;(Feldstein,1999),chapter 6;(Zweifel and Breyer,1997),chapter6.Because moral hazard and risk avoidance are both estimable,one can use this framework to estimate optimal demand prices or an optimal insurance contract.Indeed,many years ago my colleagues and I did just that as part of the RAND health insurance experiment.Specifically,we estimated that an optimal insurance policy would have an individual deductible of around$200ଝPresidential address to the2006meeting of the American Society of Health Economists.E-mail address:joseph newhouse@.0167-6296/$–see front matter©2006Elsevier B.V.All rights reserved.doi:10.1016/j.jhealeco.2006.06.0031006J.P.Newhouse/Journal of Health Economics25(2006)1005–1014in1983dollars,followed by about25%coinsurance,to a stop loss of$1500(Newhouse and the Health Insurance Experiment Group,1993),Table4.21.Our estimates were marginally higher than values common in then prevalent indemnity policies,but they did not account for the tax subsidy to employer paid premiums,which would have reduced the optimal cost sharing,nor did they account for any cashflow problems that some liquidity constrained individuals may have, which would have also reduced it.Cutting in the other direction,the estimates also assumed actuarily fair insurance.Although the RAND Experiment results were widely disseminated,after their publication the initial cost sharing in health insurance did not keep up with the increase in prices and total spending on medical care.Instead,we moved from the era of indemnity insurance to the era of managed care,typically with copayments at a level well below the foregoing calculation of an optimal insurance policy.The assumption was that supply-side cost sharing altered the tradeoff between moral hazard and risk,which it probably did,but at least two problems remained:(1) supply-side cost sharing was not well placed to affect whether consumers initiated treatment for an episode,whereas the main effect of demand-side cost sharing was on the margin of patients’initiating treatment(Newhouse and the Health Insurance Experiment Group,1993);hence,one source of moral hazard remained;(2)supply-side cost sharing by itself left no mechanism for consumers to reveal their willingness-to-pay for treatment(Pauly and Ramsey,1999;Rosenthal and Newhouse,2002).As a result,if highly valued services were not delivered because of supply-side cost sharing,it is logically possible that consumer surplus net of deadweight loss could have fallen as supply-side cost sharing rose.In any event,the possibility that some highly valued services might not be delivered because of supply-side cost sharing may have been one cause of the managed care backlash of the mid-1990s.Whether because of that backlash or because managed care was mainly about once-and-for-all reductions in spending,or both,since the late1990s real health care spending growth has resumed at approximately its historical rate.As a result,though controversial,the issue of greater cost shar-ing has come back into the discussion,most prominently with Health Savings Accounts(HSAs) and Health Reimbursement Accounts(Cogan et al.,2005;Davis,2004;Lee and Zapert,2005).The2003legislation authorizing HSAs and the regulations implementing them are reasonably prescriptive.For a policy to qualify for favorable tax treatment,the minimum deductible for an individual policy in2006must be$1050and the maximum out-of-pocket expense$5250. Preventive services are allowed to be exempt from the deductible,however,and the Treasury has defined the following services as preventive:periodic health examinations;routine prenatal and well-child care;immunizations;tobacco cessation programs;weight loss programs;and various screening programs.Importantly,however,in its initial guidance the Treasury explicitly said preventive care did not include services that treat existing conditions and so such services must be subject to cost sharing.1But is this cost sharing structure optimal?Nominal per capita health spending in2006is about five times greater than in1983.2If one is willing to ignore the differences in product over the two decades and simply uses afigure offive to adjust the optimal deductible calculated from the RAND experiment data,the$1050figure for health savings accounts is in line.The stop loss figure calculated by inflating the RAND$1500figure byfive is above the HSA maximum,but just inflating the1983dollarfigure ignores the increased risk from the increased share of income 1/offices/public-affairs/hsa/pdf/notice2004-23.pdf.2/NationalHealthExpendData/02NationalHealthAccountsHistorical.asp#TopOfPage.J.P.Newhouse/Journal of Health Economics25(2006)1005–10141007 going to health care today and hence overstates the optimal stop loss feature.It also ignores the increased number of individuals who will be liquidity constrained.Although the minimum deductible for an HSA seems roughly in line with crudely inflating the calculations from the RAND Experiment and the maximum out-of-pocket may not be too far below them,a possible improvement,consistent with the allowable exemption of preventive services,would be reduced cost sharing for chronic maintenance drugs or other medical services that treat existing conditions but could lower future expenditure,at least on expectation,and/or improve future health.I explore this possibility in the remainder of the essay.1.Empirical evidence on cost sharing and complianceCompliance with prescribed regimens is a general problem in medicine,but I am interested in a subset of the general problem,the elasticity of compliance with respect to the demand price and the effects of the resulting improved compliance on overall spending and health.Considerable evidence suggests that compliance with drug regimens by patients with certain chronic diseases increases with lower cost sharing,and it seems reasonable that improved compli-ance might reduce total lifetime spending and improve health outcomes.Hsu et al.,find that caps on drug benefits for those over the age of65reduce compliance with recommended behavior and thereby reduce drug spending,but also worsen physiologic measures of health and raise spend-ing for other medical services,even over a period as short as a year(Hsu et al.,2006).In their study the increase in non-drug medical costs,largely increased emergency department visits and increased non-elective hospitalizations,is roughly the same magnitude as the induced decrease in drug costs,but could well be greater as time passes,assuming compliance does not improve and physiologic health worsens.In earlier work Soumerai et al.showed that limiting the number of prescriptions in a month in a Medicaid program to three saved money on drugs but led to offsetting increases elsewhere;in the case of schizophrenic patients,the increase in other spending was17times the savings on drugs (Soumerai et al.,1994;Soumerai et al.,1991).(Tamblyn et al.,2001)have shown that imposition of cost sharing for drugs in Quebec decreased use of“essential”drugs among the elderly and among welfare recipients by15–22%,which led to an approximate doubling of serious adverse events,defined as hospitalization,nursing home admission,or death,as well as an increase in emergency department visits.Although they do not present cost data,the magnitude of the increase in serious adverse events seems likely to have raised total costs.(Rosen et al.,2005)have simulated that it would save money and increase QALYs for Medicare to make ACE inhibitors available at no charge for elderly diabetics when compared with the default cost sharing in the Medicare drug benefit;the induced increase in the use and cost of ACE inhibitors would be more than offset in averted medical costs.(Huskamp et al.,2003)and (Goldman et al.,2004)have shown that medications known to be efficacious for certain chronic disease,such as statins,ACE inhibitors,anti-hypertensives,and anti-diabetics,are sensitive to the amount of cost sharing,although they have not quantified the effect on other spending.An extreme version of what I have in mind–in the sense that the demand price is negative–is paying those with tuberculosis to come in so they can be observed swallowing their prescribed medications (Directly Observed Therapy).To sum up,these empiricalfindings are strongly suggestive that some individuals,especially those with certain chronic diseases,can be induced by less cost sharing to take actions today that ex ante will reduce their future use of medical services and/or improve their future health.1008J.P.Newhouse/Journal of Health Economics25(2006)1005–10142.Optimal cost sharing and compliance:elaborating the standard modelWith thesefindings as background,I want to argue that the textbook treatment of moral hazard can usefully be elaborated.In a too little noticed paper Held and Pauly considered an analogous problem,which they termed benign moral hazard(Held and Pauly,1990).Held and Pauly,however, took insurance coverage of certain medical services such as hospital and physician as given and asked about optimal coverage for“preventive”services.In principle,one must consider the optimal cost sharing structure for all services;that is,the nature of the insurance contract.In other words, there are several relevant medical“products”being insured,and one needs to consider the cross-price elasticities among them.For concreteness,however,I will stay with the example of a drug for a chronic disease such as diabetes or hypertension.An easy case is when the marginal social cost of another pill is negligible,as it often is.In that instance,assuming there is no adverse effect on the use of other medical services,a subsidy that induces greater consumption of drugs should not even be considered moral hazard,so there is no tradeoff between moral hazard and risk avoidance.But for the remainder of the paper I want to assume the additional drugs or services have non-trivial marginal social costs.And even in the case of the pill,there will virtually always be a non-trivial marginal private cost to the employer or purchaser.Before proceeding past the easy case of negligible marginal social cost,some remarks about the determinants of cost sharing are in order.Because almost all Americans with insur-ance are insured either through their employer or through a public program,the cost sharing faced by many individuals is exogenous or nearly exogenous,because it is determined by the employer or by a public body.3Furthermore,and importantly for my story,because of a com-mon insurance pool,each individual’s spending represents an external expense for others in the pool.Given the prevalence of employment based insurance,it is natural to ask what determines the employer’s decision with respect to cost sharing.(Goldstein and Pauly,1976)in their sem-inal work on the problem suggested that the employer acts to minimize labor cost and thus look at the outcome for the marginal employee.Goldstein and Pauly also discuss unions as agents for groups of employees and assume that if the union determined fringe benefits,it did so to satisfy the median employee.Although Goldstein and Pauly do not consider public insur-ance,their model of union choice would appear to extend straightforwardly to a median voter model.In a recent paper Nolan Miller has suggested that when offering two or more plans,an employer interested in minimizing total compensation cost would offer a moderate health plan,with a more generous plan available as an“upgrade”(Miller,2005).In Miller’s model the employer tries3Complete exogeneity requires several assumptions that are unlikely to hold completely but may hold approximately. One is that individuals do not choose their employer on the basis of the amount of cost sharing in the health insurance plan(s)the employer offers.A second is that the individual either is single or that the spouse’s employer does not offer insurance with materially different cost sharing or that the spouse does not work.The clearest case of exogeneity or near exogeneity occurs when an employee has no choice of plan at work and the employer subsidizes the plan sufficiently that it is not rational to decline coverage.But even for those who have a choice among plans at the workplace,the employer may have standardized cost sharing among the options.For many Medicare beneficiaries the only choice of cost sharing within traditional Medicare is through the purchase of an individual Medigap policy,which only a minority of beneficiaries choose,perhaps because of high loading charges,though others may have a choice about the purchase of supplementary insurance through the employer.I think one can safely ignore the possibility that an individual chooses a job based on the availability of retiree health insurance.J.P.Newhouse/Journal of Health Economics25(2006)1005–10141009 to manage selection among multiple plans to minimize total compensation cost.4In effect,the employer has some market power in selling insurance to workers and prices the upgraded policy to recover some of the surplus workers receive.Assume a multi-period context,that the incidence of employer-paid premiums is on employees, and that the employer is minimizing cost.Then the employer would structure cost sharing for any given current level of overall employer subsidy to induce individuals in the group to act so as to minimize the group’s lifetime health care costs for the period of time they are in the group and would also account for any effects on productivity at the workplace.5In this case if subsidizing drugs for chronic diseases or routine maintenance visits improves compliance and reduces health care costs for the group,the increased consumption is not a deadweight loss,as the usual textbook model suggests,but rather a subsidy to recognize the positive externality within the group.Ceteris paribus,that externality will be greater,the longer job tenure is with an employer. It follows that a cost-minimizing employer will provide more generous insurance for medical services that decrease future health care insurance costs and/or increase future productivity,the longer the job tenure at that employer.I leave it for others to test this implication.If the employer does not consider benefits after the person leaves employment,for example savings to Medicare, cost sharing for these types of services will be too high from a social perspective.I note in passing that when calculating the moral hazard-risk avoidance tradeoff,the substantial intertemporal correlation in spending among those with chronic disease means the calculation of risk on a lifetime basis,holding constant cost sharing,will be greater than on an annual basis. This caveat applies to the calculations from the RAND Experiment data on the optimal tradeoff between moral hazard and risk avoidance cited above.3.Optimal cost sharing and compliance:a behavioral economics modelThe foregoing is an elaboration of traditional theory,but I think there is more to the story. One might ask why an individual does not comply with a physician’s recommendations to take medication even with no subsidy.A standard answer,which surely applies in a number of cases, points to the insured nature of any medical costs resulting from non-compliance and argues that the private monetary benefits of compliance are not that large.And perhaps the non-monetary benefits are not all that large either if the problem is not very annoying.Or,perhaps more likely, there could be side effects that outweigh the benefits.But this answer does not seem very plausible for many of the chronic diseases considered above.We probably all heard as children the injunction from our mothers to“take your medicine.”Our mother’s“advice”was probably delivered with a tone of“you will take your medicine”rather than“you should take your medicine,”suggesting we might not have taken it if our mothers had not been around to admonish us.This implies some greater inducement than simply information may be needed to induce compliance that will improve our future health–and possibly save us money depending on the structure of cost sharing for other medical services–and that a model from behavioral economics might be applicable.64In the usual case in which wages are not adjusted for individual health costs,the cost minimizing solution will not be an equal subsidy to all plans offered.5If there is an effect on pensions,in principle the employer would account for that as well.6This case is somewhat analogous to those cases considered by della Vigna and Malmendier,who show that if con-sumption of a good today raises future demand,a profit-maximizingfirm will price the good today below marginal cost, and conversely(della Vigna and Malmendier,2004).1010J.P.Newhouse/Journal of Health Economics25(2006)1005–1014 Beginning with(Strotz,1955–1956),economists have worked on models with time-inconsistent preferences,or discount rates that are higher in the short run than the long run.A common example of behavior that these models generate comes from individuals who put off saving or dieting today but plan to begin tomorrow;when tomorrow comes,however,these per-sons decide to put it off another day,and so on.In other words,the assumption of time-inconsistent preferences manifests itself as a self-control problem.There is now a considerable theoretical literature in this domain,including among others (Phelps and Pollak,1968),(Thaler and Shefrin,1981),(Ainslie,1991),(Akerlof,1991)(Laibson, 1997),and(O’Donoghue and Rabin,1999a,b).There is also a large amount of experimental evidence from psychologists that supports time-inconsistent preferences;see(Ainslie,1992),as well as references in several of the papers just cited.7An excellent review of this and much additional literature is(Frederick et al.,2002).To write down a model,I use Laibson’s model of a utility-maximizing consumer with quasi-hyperbolic discounting;i.e.,a discount rate on immediate decisions that exceeds that for decisions in the future,with a constant discount rate on future decisions.In a conventional health economics model with time-constant exponential discounting,utility at a point in time t is a function of the consumption of non-medical goods and services x and a stock of health H t,which in turn depends on the consumption of x as well as medical goods and services m.Thus,U t=U(x t,H t(m,x)),(1) where I have not subscripted either m or x in the health care production function to indicate that health may depend upon prior period consumption.Discounted lifetime utility in the standard model is:T−ti=0δi U t+i(2)Individuals are assumed to act to maximize(2)subject to a lifetime budget constraint.In some formulations of the model x and H are also functions of time spent in the activity,and there is a time budget constraint as well as an income constraint.Importantly for my purposes,in this model the assumption of time-consistent preferences implies that if I prefer to wait one day to undertake an activity today,I will prefer to wait one day before undertaking the activity two weeks from now,and conversely.In the quasi-hyperbolic discounting model discounted lifetime utility is instead:U t+βT−t1δi U t+i(3)δandβare between zero and one,withβstrictly less than one,so that there is a discontinuity in the discount rate between the initial period(δ)and all subsequent periods(βδ).8 Solutions to maximizing(3)subject to a lifetime budget constraint depend upon assumptions about the awareness of the individual of the time-inconsistent preferences.One extreme assump-7Work in neuroeconomics suggests time-inconsistent behavior may have a biological basis,in that different parts of the brain process behavior dealing with immediate gratification or avoidance of harm than behavior consistent with rational calculation of future consequences(Bernheim and Rangel,2004).8In more general forms of the model the discount rate declines smoothly as a function of the distance from the present.J.P.Newhouse/Journal of Health Economics25(2006)1005–10141011 tion is that the individual is myopic and does not realize that he or she has such preferences (na¨ıve behavior),and therefore does not account for future behavior in today’s actions.The other extreme assumption is that the individual knows that he or she has such preferences and therefore accounts for those preferences in current behavior(sophisticated behavior).In general,solutions are easier if na¨ıve behavior is assumed.Furthermore,solutions can differ depending whether the action in question generates immedi-ate costs or immediate rewards.In the compliance application I have in mind there are immediate costs,in which case a small bias for the present–βonly a little less thanδ–can generate a large welfare loss if there is na¨ıve behavior(O’Donoghue and Rabin,1999a).Recently these models have been applied to cigarette smoking by Gruber and Koszegi(Gruber and Koszegi,2001,2004),with the important results that:(a)if individuals have time-inconsistent preferences,welfare would be maximized with much higher cigarette taxes than at present because health benefits would accrue to individuals who would otherwise begin to smoke;in effect,the higher taxes induce individuals to comply with a no smoking regime;(b)because the benefits of not beginning to smoke accrue disproportionately to lower income individuals,cigarette excise taxes are much less regressive than usually assumed and may even be progressive.In addition to applications involving addictive behavior,the model of time-inconsistency sketched above can apply straightforwardly in the context of lowering cost sharing to induce greater compliance.9In effect,there is a subsidy for something with a future good rather than a tax for something with a future bad.In particular,there may well be cases in which subsidizing certain medical goods or services increases total medical cost,but increases health sufficiently such that the benefits outweigh the cost increase.In effect,the individual may overweight imme-diate costs relative to future benefits.From a social perspective,thefirst order condition in this case would decrease cost sharing until the marginal health benefit equaled the marginal increase in induced health care cost.The type of cost sharing structure suggested by the empirical studies above–selected subsidies for specific medical services for specific persons,such as the full coverage of ACE inhibitors for diabetics suggested by(Rosen et al.,2005)–presumably are not observed for reasons of selection. The recent growth of disease management and case management,however,seems to be a non-price method intended to accomplish similar ends by targeting specific individuals for specific medical interventions.A cost-minimizing employer,of course,would not lower cost sharing for medical goods that are not valued by the employee at the present.An employer who has an element of paternalism or a public plan such as Medicare,however,might do so.Inducing greater compliance with such medical goods as chronic maintenance drugs is in the spirit of Save More Tomorrow,a method several employers now use to increase employee saving by having employees commit today to earmark a portion of future raises to401(k)plans(Benartzi and Thaler,2004).It is also in the spirit of the action by some employers to make the default in their401(k)plans enrollment in a prespecified fund or combination of funds rather than no enrollment(Madrian and Shea, 2001).Importantly,however,in these last two examples the employer’s actions are costless to the employer,whereas more complete coverage of certain drugs or services is potentially costly.David Cutler has suggested to me a few additional applications of behavioral economics that an employer might apply to induce healthier behaviors.One is analogous to the Save More Tomorrow9The use of models with time inconsistency is but one use of behavioral economics in health economics;other appli-cations of behavioral economics to health care are discussed in an insightful paper by Richard Frank(Frank,2006).1012J.P.Newhouse/Journal of Health Economics25(2006)1005–1014idea;employers could have employees who smoke pay something today and for the next several months.If they completed a smoking cessation course in the future,say next year;the monies they paid would be refunded with an additional amount.A second is to lower the time cost as well as the money cost of healthy behaviors and raise the time cost for unhealthy behaviors.Discovering opportunities to modify the health insurance contract by subsidizing specific medical goods or services in order to minimize future medical costs and/or improve health and productivity seems like a rich vein for empirical research,including experimental research,even though past research has not been kind to suggestions that coverage of certain services would reduce medical costs.I offer two examples.Despite evidence from observational data that it would do so(Roemer et al.,1975),the RAND Experiment data showed that more complete coverage of office visits did not reduce future costs by detecting disease early but instead increased both inpatient and outpatient costs.10The Channeling demonstration showed that increased coverage of home and community based services only modestly reduced the use of nursing home services and on balance increased costs,though it did produce benefits(Kemper,1988).I believe the problem in these instances were that the subsidies were not sufficiently specific;rather than a subsidy for all office visits,for example,there could be a subsidy,or an additional subsidy,for routine maintenance visits for those with certain chronic diseases.And because some of the cost-saving benefits from subsidies in employment-based insurance may spill over to Medicare, whereas others may spill over to future employers that provide health insurance,strong evidence of cost saving would provide a case for certain mandates.To sum up,from a social perspective there are reasons from both the nature of health insurance markets and individual behavior to suspect that markets will not lead to optimal cost sharing. Employers may not internalize future cost reductions,and consumers may systematically under-value future benefits.Moreover,inducing greater consumption of some drugs may not even represent moral hazard in the conventional sense of the term if the social cost of another pill is negligible.Even from a private perspective,uniform cost sharing across people within a group and across services such as physician visits or drugs is unlikely to be optimal.If it was,the market for disease management services would be limited.Discovering opportunities to refine cost sharing seems like an important part of the health economics research agenda.In any event,optimal cost sharing looks to be much more specific than the current statute law and resulting regulations mandates for HSAs.AcknowledgementsI would like to thank David Cutler,Richard Frank,Jon Gruber,Tom McGuire,and Richard Zeckhauser for comments on a draft of this paper.The usual disclaimer applies.ReferencesAinslie,George,1991.Derivation of‘rational’economic behavior from hyperbolic discount curves.American Economic Review81(2),334–340.Ainslie,George,1992.Picoeconomics:The Strategic Interaction of Successive Motivational States.Cambridge University Press,New York.Akerlof,George,1991.Procrastination and obedience.American Economic Review81(2),1–19.10See(Helms et al.,1978)for the problems with the observational data.。