Homework answers / question archive / O v e rv i e w Is Health Spending Excessive? If So, What Can We Do About It? The delicate task of reining in spending without harming our welfare

O v e rv i e w Is Health Spending Excessive? If So, What Can We Do About It? The delicate task of reining in spending without harming our welfare

Economics

Share With

O v e rv i e w Is Health Spending Excessive? If So, What Can We Do About It? The delicate task of reining in spending without harming our welfare. by Henry J. Aaron and Paul B. Ginsburg ABSTRACT: The case that the United States spends more than is optimal on health care is overwhelming. But identifying reasons for excessive spending is not the same as showing how to wring it out in ways that increase welfare. To lower spending without lowering net welfare, it is necessary to identify what procedures are effective at reasonable cost, to develop protocols that enable providers to identify in advance patients in whom expected benefits of treatment are lower than costs, to design incentives that encourage providers to act on those protocols, and to provide research support to maintain the flow of beneficial innovations. [Health Aff (Millwood). 2009;28(5):1260–75; 10.1377/hlthaff.28.5.1260] A c h i l d t r a i n e d to s ay “ c o s t, q ua l i t y, ac c e s s ” might pass as a health policy analyst. Sustaining the deception would become more difficult if the innocent were asked to define those terms. “Access” would not pose much problem: being able to obtain care when you believe you need it. Defining “quality” would be more challenging. In concept, it means that patients are getting the outcomes expected from the application of current medical knowledge. Often, however, the practical difficulties in measuring outcomes and adjusting risk variation among patient populations means that “good” quality is instead defined in terms of whether certain processes have been followed. The problem of rising cost—or, more accurately, spending—once again, seems clear enough. But is it? To begin with, spending is simply price times quantity. Is the problem excessive price, excessive quantity, or both? Those troubled by rising spending seem to have a range of concepts in mind (see Exhibit 1). If spending is rising and if that seems problematic, the practical questions are as follows: what exactly is wrong with spending more on some good than one spent in the past? And what tools are available to control spending on something that is beneficial on average but not for each patient? Level Of Health Spending n U.S. per capita spending. The United States spends more per person on Henry Aaron (haaron@brookings.edu) is the Bruce and Virginia MacLaury Senior Fellow, Economic Studies, at the Brookings Institution in Washington, D.C. Paul Ginsburg is president of the Center for Studying Health System Change, also in Washington. 1260 September/October 2009 DOI 10.1377/hlthaff.28.5.1260 ©2009 Project HOPE–The People-to-People Health Foundation, Inc. Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng EXHIBIT 1 Commonly Asked Questions About Health Care Spending Question Answer Level Does the U.S. spend more per capita than other countries? Does the U.S. spend more per capita than other countries after adjusting for income and other factors? Yes, a lot more Yes, a lot more Growth Has per capita U.S. health care spending risen faster than that of other countries in recent years? Has the excess of health spending growth over income growth been higher in the U.S. than elsewhere in recent years? Waste Does the U.S. spend a lot on low- or no-benefit care? Does the U.S. spend more on low- or no-benefit care than other countries? Price Does the U.S. pay higher prices for health care services than other countries? Does the U.S. pay more for health care services, adjusting for quality, than other countries? Welfare Would cutting (growth of) health care spending raise welfare? Fiscal issues Will increases in health care spending impose stress on public budgets? More than some, less than others In general, yes, but by varying amounts and not uniformly Yes, but evidence on how much is poor We don’t know, but our larger outlays mean that we could waste more Definitely Almost certainly, although measuring—and even defining—quality is difficult; in some dimensions, we seem to be doing very well (length-of-stay, cutting-edge procedures); in others, poorly (delivering recommended care, control of diabetes); we have few data from other countries Static: if one could target cuts, yes; if not, no Dynamic: if cost limits improve targeting of research, yes; if not, could be harmful Yes; were it not for projected increases in health care spending, no material long-term gap between revenues and expenditures under current policy would exist SOURCE: Authors’ synthesis. health care than does any other nation—roughly twice the average of the ten richest countries other than the United States.1 Spending more than others do on health means spending less than others do on other private or public services—education, housing, income security, or national defense, for example. Several factors contribute to this high level of spending. First, health care spending rises with income, and U.S. per capita income is higher than that of most other nations.2 However, neither the level nor the growth of per capita income can explain why U.S. spending is so much higher than that of other nations or why it has grown so fast. Even after one controls for national differences in per capita income (and for the range of other fac- H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1261 O v e rv i e w tors that are thought to influence health spending, such as the average age of the population), U.S. per capita health spending is about 30–40 percent above what income and other factors can explain.3, 4 Furthermore, the gap between per capita health spending in the United States and that in other developed countries bears little relation to the relative growth of per capita income (Exhibit 2). n Health spending and outcomes. Per capita health spending varies widely among countries and among the various U.S. states. In general, simple correlations indicate that there is little or no connection between health spending and both life expectancy and infant mortality, whether one is comparing developed nations or U.S. states (Exhibits 3 and 4). Multivariate analyses that include such additional determinants of health as incomes, environmental quality, and personal habits do not change this conclusion.5, 6 The connection between spending and health outcomes could be loose for several reasons. One could be that health care is not an important determinant of health outcomes; however, several studies suggest that this conclusion is false.7–12 Another might be that areas that deliver technologically sophisticated—and costly—care are inefficient in delivering less-sophisticated care.13 A third reason might be that much health spending goes to relieve conditions, such as joint deterioration, cataracts, and some forms of angina, that cause disability, not death. Whatever the reason, it is hard to avoid the conclusion that the United States is buying less health than other nations do with its high outlays. n Reasons for high U.S. health spending. Much of the excess of U.S. spending is attributable to the fact that the unit prices of various services are higher in the United States than elsewhere. Some part of the high prices goes to incomes of highly trained personnel. But in some cases, such as outpatient services, much of the price EXHIBIT 2 Growth Of Per Capita Income And Health Spending, United States And Twenty Organization For Economic Cooperation And Development (OECD) Countries, 19702006 United States Twenty OECD countries Income growth (percent) Excess growth of health care spending over income (percent per year) Period Income growth (percent per year) Excess growth of health care spending over income (percent per year) 1970–1980 1980–1990 1990–2000 2000–2006 2.20 2.28 2.04 1.41 2.22 3.11 1.07 2.49 2.54 2.20 2.04 1.60 3.10 0.80 1.21 1.89 1970–2006 2.05 2.09 2.11 1.66 SOURCE: OECD Health Data 2008. Version 12/10/2008. Available from http://fiordiliji.sourceoecd.org/v1=6937783/c1=34/ nw=1/rpsv/statistic/s37_about.htm?jnlissn=99991012 NOTES: The countries (and the periods covered) included are Australia (1971–2006), Austria, Belgium, Canada, Denmark (1971–2006), Finland, France, Germany, Ireland, Italy (1988–2006), Japan, the Netherlands (1972–2004), New Zealand (1970–2003), Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. 1262 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng EXHIBIT 3 Total Health Spending Per Capita Compared To Life Expectancy At Birth, United States (By State) And Nineteen Organization For Economic Cooperation And Development (OECD) Countries, Various Years Age (years) OECD countries 82 United States 80 78 76 74 72 1,000 2,000 3,000 4,000 5,000 6,000 Per capita health spending (dollars at U.S. exchange rate) 7,000 SOURCES: OECD countries: OECD health data 2008. Paris: OECD; 2008 Dec. [cited 2009 Mar 25]; via SourceOECD. U.S. health spending: Henry J. Kaiser Family Foundation. Health care expenditures per capita by state of residence, 2004 [Internet]. Menlo Park (CA): Kaiser Family Foundation; 2004 [cited 2009 Mar 25]. Available from: http://www.statehealthfacts.org/comparetable .jsp?ind=596&cat=5&sub=143&yr=14&typ=4&sort=a. U.S. life expectancy: U.S. Census Bureau. U.S. projections methodology table no. 2: average life expectancy at birth by state for 2000 and ratio of estimates and projections of deaths: 2001 to 2003. Washington (DC): U.S. Census Bureau, Population Division, Interim State Population Projections; 2005 Apr 21 [cited 2009 Mar 25]. Available from: http://www.census.gov/population/www/projections/methodology.html EXHIBIT 4 Total Health Spending Per Capita Compared To Infant Mortality (Deaths Per 1,000 Live Births), United States (By State) And Eighteen Organization For Economic Cooperation And Development (OECD) Countries, Various Years Deaths per thousand live births United States 10 8 OECD countries 6 4 2 1,000 2,000 3,000 4,000 5,000 6,000 Per capita health spending (dollars at U.S. exchange rate) 7,000 SOURCES: OECD countries: OECD health data 2008. Paris: OECD; 2008 Dec. [cited 2009 Mar 25]; via SourceOECD. U.S. health spending: Henry J. Kaiser Family Foundation. Health care expenditures per capita by state of residence, 2004 [Internet]. Menlo Park (CA): Kaiser Family Foundation; 2004 [cited 2009 Mar 25]. Available from: http://www.statehealthfacts.org/comparetable .jsp?ind=596&cat=5&sub=143&yr=14&typ=4&sort=a. U.S. infant mortality: Kaiser Family Foundation. Infant mortality rate (deaths per 1,000 live births), linked files, 2003–2005 [Internet]. Menlo Park (CA): Kaiser Family Foundation; [cited 2009 Mar 25]. Available from: http://www.statehealthfacts.org/comparetable.jsp?ind=47&cat=2&sub=13&st=3&yr=79&typ=3&sort=a H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1263 O v e rv i e w difference goes to support inefficient production made possible by a lack of competition or effective regulation.4 High prices sometimes serve as a proxy for high quality—more or better equipment or better-trained personnel. We know of no hard evidence showing that the quality of high-price U.S. services is better than that of corresponding services elsewhere or whether and to what degree it accounts for higher U.S. prices. In some cases, however, price differences are so large (for example, magnetic resonance imaging studies in the United States and Japan)14 that no plausible quality difference can explain the gap. Growth Of Spending For decades, health spending has claimed a growing share of national income in the United States and in most other nations. Tautologically, that increase occurred because health spending grew faster than income. But the excess, which—also tautologically—depends on the difference between the growth rates of health spending and of income, also varies widely. The gap has tended to be larger in the United States than in most other nations, but not at all times (Exhibit 2). Most studies attribute one-half to two-thirds of the gap to the advance of medical technology, which lengthens the menu of beneficial interventions or improves their quality.2 It is doubtful whether similar studies based on data from other nations would yield exactly the same results, given the large differences among countries in the growth of health spending and per capita income. Furthermore, projections indicate that health spending will continue to claim a growing share of U.S. income.15–17 In the past, population aging accounted for little of that growth. In the future, it will contribute a larger but still modest amount— about 0.4 percentage points per year. But that is less than one-fifth of the projected gap between health spending and income trends.2 Although population aging explains little of the projected increase in total health spending, increased spending on care for the elderly and disabled is expected to become a severe fiscal burden for the U.S. federal and state governments. The Congressional Budget Office (CBO) projects that the share of gross domestic product (GDP) devoted to Medicare and Medicaid will approximately quintuple between 2009 and 2050.16, 18 Were the gap between growth in health care spending and income to persist, nonhealth consumption for the working-age population would eventually decline.17 Although it is difficult to speculate about reactions to the stresses of a situation in which health care spending growth crowds out spending for other goods and services to such an extent that it reduces them, it is already apparent that the strains of health spending growth exceeding income growth in recent years are falling more heavily on lower-income people through erosion of private health insurance coverage and the financial burdens of care for those with less comprehensive insurance or who no longer can afford insurance. 1264 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng What Does Higher U.S. Health Spending Buy? Exactly why Americans spend so much on health care is not well understood. Some unknown proportion of higher U.S. spending supports economic rents— payments larger than necessary to keep health care resources in their current use. Some goes for superior quality. Some goes for low- or no-benefit services. Some results from inefficient production methods, including wasteful spending. Administrative complexity has been much studied, but how much administrative spending is wasteful and how much it may contribute to the growth of overall health care spending remain controversial.19, 20 More generally, some commentators have alleged that as much as one-third of all health spending is wasteful.21 These studies extrapolate to all health spending findings from research based on treatment of individual diseases. Some studies indicating waste have been modified or reversed by subsequent research.22 Furthermore, the very definition of waste is rarely specified.23 A key assumption lies behind assertions of waste—that information exists on how, ex ante, to distinguish care that is worth what it costs from care that is not. This assumption is often untrue. Much evidence indicates the misallocation of health spending. For example, the expected cost of adding a quality-adjusted life-year (QALY) varies enormously among widely used medical interventions.24 That finding, which is consistent with any of the explanations for higher U.S. spending, suggests that reallocation of spending could improve outcomes. So does evidence from the Dartmouth Atlas of Health Care25 documenting that variations in the use of many procedures are out of all proportion to any conceivable differences in the incidence of illness or “tastes” of patients for different methods of treatment. Various studies have reported estimates of the aggregate value of health spending in terms of increased longevity and reduced sickness that are traceable to increases in health spending. These studies indicate that measured over long periods, the value of better health outcomes for selected conditions that seem traceable to improved treatments exceeds increases in total health spending.7–12 But these studies are for selected improvements. Also, the authors all acknowledge that their findings are fully consistent with other research indicating waste or inefficiency on the margin, that the margin can be wide, and that spending within that margin is large. For example, interventional cardiology procedures have been shown to have very high value in some patients, such as those who have had a heart attack, but recent research raises doubts about the value of such procedures for asymptomatic patients who have worrisome results from diagnostic testing.26, 27 Furthermore, there is some reason to think that the ratio of the total added value of medical spending to total added cost has diminished in recent years.10, 12 H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1265 O v e rv i e w Causes And Cures The coexistence of high average returns and low marginal returns per dollar spent on health care raises a key question: how can the level of or growth in spending be lowered in ways that will enhance welfare? Exhibit 5 summarizes various influences on the level of and growth in health care spending, whether these influences lead to excessive spending levels or growth, and whether policy changes to curtail spending would increase welfare. EXHIBIT 5 Factors Contributing To Excessive Level Of Or Growth In Health Spending Factor/ influence Does this factor raise spending? Can or should policy changes affecting this variable reduce growth of health care spending? Does this factor cause excessive spending? Does this factor raise spending growth? Does this factor make spending grow excessively? Not clear; insurance increases both lowand high-value services (such as prevention and adherence to drug regimens) Yes; when coverage is deep, likely stimulates more investment in development of technology Yes (unless curbed by other means); overpayment for services based on newer technology leads to excessive spending growth Yes, through tax system (see below) Demand Insurance Yes—as indicated by theory (moral hazard) and empirical research (RAND) Tax system Yes, by encouraging Probably, although insurance impact on net benefit unclear No direct impact unless coverage increases; insurance may promote costincreasing technological change Unclear; hinges on whether induced technological change, on balance, is worth more than it costs Yes; cap exclusion or replace with refundable, capped credit Income Yes No reason to think so Yes, if income elasticity is >1 No reason to think so No Yes, whether or not physicians act as perfect agents; added costs for administration Yes, whether or not physicians act as perfect agents Unclear; yes, if payment system distorts research incentives Probably Yes; somewhere between episodebased reimbursement and capitation; either can be blended with fee-for-service No No; possible shortages emerging Perhaps, if it intensifies technological arms race Probably; increase primary care and use of non-MDs Supply Fee-for-service payment system No. of providers Unclear: MDinduced demand versus competition over price Provider mix (MDs vs. other health professionals; specialists vs. GPs) 1266 Yes, to the extent of No MD-induced demand Probably, unless Specialty mix probably pushes up specialist quality offsets added cost spending; limited delegation to nonMDs probably leads to higher spending, except for induceddemand offsets Unclear, too fast if professional mix induces development of low-benefit, specialty-intensive procedures and devices September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng EXHIBIT 5 Factors Contributing To Excessive Level Of Or Growth In Health Spending (cont.) Factor/ influence Does this factor raise spending? Does this factor cause excessive spending? Organization of hospitals and physicians offices Fragmentation likely Yes, through leads to higher inefficiencies from spending fragmentation and from suboptimal scale of smaller practices and hospitals Managed care Managed care at least lowers prices; HMOs (prepaid groups) lower spending level No Does this factor raise spending growth? Does this factor make spending grow excessively? Can or should policy changes affecting this variable reduce growth of health care spending? Yes, “medical arms race” Fragmentation limits potential for productivity gains over time and may slow development and application of new cost-saving techniques Payment reforms that reward organizations for lower costs per episode or per capita and that reward higher quality May slow take-up of costly new technology No No No Institutional factors Litigation Adversarial legal Small impact: system generates premiums negligible, defensive dead-weight costs medicine hard to measure but not likely to be large Pay levels, drug Large effect; and device fragmentation of prices U.S. payers (except Medicare, which cannot use its powers) Yes, excess prices (rents) Unclear; data do not Yes; costs of some support firm products and judgment services are excessive Patent system Not clear; question is whether the design of the system could be improved and how one would know Encourages research, the key driver of increasing health spending Encourages research, the key driver of increasing health spending Not clear; question is whether the design of the system could be improved and how one would know Move to nonadversarial compensation system Antitrust policy, rate setting Prevent current patent holders from blocking innovation when patents expire Research Equipment and Highly cost procedures increasing In some cases; but Yes benefits of medical advances exceed cost, on the average Selectively, but most is worth the added cost on the average Drugs Short-term reductions in costs; long-term, generally cost increasing, although that may change Currently negative, No, except that some drugs enable because of empty low-benefit terminal pipeline (cancer) treatment Large outlays to produce slight tweaks on existing drugs are wasteful; overall, drug innovation produces benefits far in excess of cost Comparative effectiveness, cost-effectiveness Little effect; there is To the extent that it is done, no so little of it, large potential not realized To the extent that it is done, no Target spending should depend on research opportunities No effect at present; Essential for rational could lower growth decisions on or improve targeting curtailing coverage or services SOURCE: Authors’ analysis. H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1267 O v e rv i e w We grouped them in four categories: demand, supply, institutions, and research. The unsurprising verdict is that several factors contribute to excessive spending and to unduly rapid growth, but that devising ways to correct both problems in ways that promote welfare is politically challenging and technically difficult. n Demand. Insurance is intended to boost demand for health care services, preventing the direct cost from discouraging its use by people who need it. At least since the 1960s, it has been well recognized that in performing its intended function, insurance risks encouraging excessive demand for care.28, 29 The RAND Health Insurance Experiment (HIE) provided estimates of the size of the increase; although the data are more than three decades old, the results continue to be used today.30 Whether the increase in demand for health care increases or lowers welfare depends, in a narrow sense, on the balance between the tendency of insurance to encourage people to seek some care that is worth less than its total societal costs and the fact that insurance enables people to afford high-value care that would otherwise be beyond their means. Insurance also spreads risk, an important benefit, because most people are thought to be risk-averse. But insurance does more than spread the risk of unpredictable variations in health care spending. It also spreads the costs of expected (predictable) variations in use of health services based on genetic endowments, individual histories, and personal behavior. The social value of spreading the misfortune of an unlucky “draw” from the genetic pool seems clear. But spreading the risks associated with unhealthful or dangerous behavior is more difficult to defend. As noted below, insurance also shapes the way in which providers organize to supply health care services. Tax provisions—the exclusion of employer-financed health insurance from personal income and payroll taxes—shield people from the full costs of health insurance and, hence, of the health care that insurance supports. For example, taxpayers subject to the full payroll tax who are in the 25 percent personal income tax bracket pay 37.4 percent less for health insurance (in terms of the amount of other consumption goods forgone) if their employers buy the insurance for them than if they buy it for themselves. (This difference excludes state income tax benefits and any additional advantage because selling and administrative costs are lower for group than for individual insurance.) Thus, the tax exclusions encourage people to buy more health insurance and, indirectly, more health care than they would purchase if they faced the full, before-tax cost of care that is financed with insurance. Standard economic analysis suggests that these tax policies cause people to buy more insurance than is optimal, but recent work from behavioral economics raises the possibility that tax policy may offset common cognitive errors—such as overweighting near-term costs of preventive care and underweighting the value of the deferred benefits such care generates—and thereby improves welfare.31, 32 If people do buy excess insurance because of tax provisions, the static loss can be considerable, but the dynamic effects could be much larger. Excess insurance is likely to bias incentives governing medical research. If insurance blinds people to 1268 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng the cost of care they use, they will be insensitive to the relative price of different services, tending to use excessively costly care and too much of it. That behavior, in turn, will encourage research entrepreneurs to invest in innovations that are needlessly expensive in relation to their potential for improving outcomes. Furthermore, research responds to economic incentives.33 Distortion of the character of new products and services could impose much larger losses over time than does any static resource misallocation. Unfortunately, there is no quantitative evidence on the size of such distortions. The effects need not be all negative, however. Excessive use resulting from insurance encourages investments in the development of new products. Although many might not be worth what they cost, the added benefits from even some of those innovations could exceed the cost of the whole enterprise.5 n Supply. Particular aspects of the way health care is supplied in the United States result in a higher level of spending and may spur excessive growth. The aspect most commonly cited is fee-for-service reimbursement for health care services. That method of payment rewards providers for supplying particular services rather than for producing favorable outcomes or efficiently treating an episode of illness. Furthermore, it encourages physicians to provide services beyond those they would offer as well-informed and unbiased agents for their patients. n Inefficient organization. Many observers believe that inefficient organization of health care delivery needlessly boosts U.S. health spending. These alleged inefficiencies include the continuation of the single-practitioner physician office; the survival of low-occupancy and inefficiently small hospitals; and the slow adoption of modern information technology (IT) in physicians’ offices and hospitals, which precludes the use of electronic medical records and e-prescribing, hampers data collection for research on comparative effectiveness, and results in needless duplication of tests. Many of these inefficiencies are likely fostered by a fee-for-service payment system that induces providers to provide more billable services, especially those with more generous reimbursement, as opposed to rewarding efficient solutions to patients’ health problems. Advocates of integrated delivery systems, still relatively uncommon among U.S. health care providers, argue that they are capable of delivering more health care less expensively than is possible when providers are separately managed.34 But with the notable exception of Kaiser Permanente, which has completely integrated financing and delivery, the existing payment system has posed serious barriers to realizing the promise of integrated delivery.35 n Physician supply. An important question for public policy concerns how a change in the number of providers would affect health care prices, total health care spending, and public welfare. The standard—and overly simplistic—response is that boosting the number of providers will lower prices and total spending. Even if the assumption about prices is true, an increased supply of physicians will lower total spending only if physicians cannot materially increase demand for their own services and if the increased supply does not release bottlenecks in care provision. A H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1269 O v e rv i e w huge body of research has been produced to show that physician-induced demand does or does not exist.36 If it does exist, there is little evidence about how much physician-induced demand does not confer medical benefit. The impact of an increased number of providers on outlays and welfare depends not just on numbers but also on mix. For example, the proportion of U.S. physicians with specialty and subspecialty training has grown rapidly as the complexity of diagnoses and treatments has increased. In part, the growing supply of highly trained specialists is a response to advancing technology. In part, however, it results from inadvertent payment incentives. Adjustment of payment rates, initially set when procedures are new and costly, tends to lag falling supply costs, leading to excessive payment rates. Overly generous payment then fosters overuse of sophisticated interventions and attracts excessive numbers of young physicians into those specialties in which such interventions are most important. n Institutional factors. Malpractice litigation. The view that malpractice litigation raises the level and growth of health care spending is widespread. Large awards are alleged to drive up insurance premiums and practice costs. The threat of litigation is alleged to generate “defensive” medicine: the provision of low- or no-benefit care by physicians solely to minimize the likelihood that they can be successfully sued or sued at all. To reduce these costs, several states have enacted, and Congress and other states are considering, caps on compensation for noneconomic losses from medical negligence. Evidence that malpractice litigation and its threat have much impact on the level of health spending is weak and confined to a few specialties. Evidence that it perceptibly raises the growth of health spending is almost nonexistent.37 The chief shortcoming of the malpractice dispute resolution system is not that some settlements are excessive, but that transaction costs are enormous. Litigation expenses and administrative overhead absorb roughly 60 percent of premiums. Fewer than 10 percent of victims of medical negligence ever receive compensation or, indeed, even make it to court.38, 39 Reforms of the dispute-resolution system that encourage providers to reveal medical errors, emphasize mediation, and establish simplified methods of determining compensation hold the promise of increasing the share of malpractice premiums that go to injured patients rather than lawyers and expert witnesses.40 It is less clear what their impact would be on the overall cost of the dispute resolution system and hence on health spending. Physician payment. Physicians are paid more in the United States relative to average income than in any other country.4 The proportion of U.S. physicians with specialty and subspecialty training is also higher than elsewhere; and the fee-forservice payment system may encourage U.S. physicians to work longer hours than physicians elsewhere, who are often salaried or paid by capitation. Thus, part (or conceivably all) of the additional pay of U.S. physicians may be explained as compensation for additional training and may be associated with higher qualityadjusted productivity or longer working hours. Even if one concluded that U.S. 1270 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng physicians are paid too much and their compensation should be reduced, the impact of even massive cuts on health care spending would be modest. A 25 percent cut in payments for physician services over the next decade (which would imply a far larger drop in physician income, because practice expenses would not fall commensurately) would lower the projected annual growth of U.S. health care spending only from 6.2 percent to 5.7 percent.41 The key to reduced spending by physicians is not adjusting their salaries but encouraging them to use and order fewer and less complicated services—including laboratory and radiological services. Drug prices. Prices of drugs in the United States are higher than in other countries, although the exact margin is hard to pin down.4, 42 Countries tend to consume relatively more of drugs that are priced comparatively low in that country; differences in pill sizes also make comparisons tricky. Medical device prices are higher in the United States because hospitals often must purchase whatever model each physician on the staff prefers to use (so that they continue to admit patients) and are prohibited from offering physicians a share of the savings from standardizing devices to be used. Different problems afflict the pricing of durable medical equipment (DME), such as wheelchairs or walkers, which in many cases can easily be purchased for less than Medicare’s official price schedule. Despite such rigidities, efforts to institute competitive bidding under Medicare for DME foundered because of opposition from adversely affected suppliers. Patent system. The patent system is one of the most important and complex institutions affecting health care spending. A patent is a government-sanctioned monopoly that enables the developer of a new product to charge higher-thancompetitive prices for a fixed period of time. Whether patents raise spending or lower it at a point in time depends on the facts and circumstances. They raise spending if newly developed products fostered by patents replace less costly, but presumably less effective, products or if they increase the number of treatable conditions. Examples of such expenditure-increasing advances include antibiotics, which reduced treatment costs for infectious diseases but, by sparing people inexpensive deaths from infectious diseases, enabled them to die later from more costly conditions. Other cost-increasing advances include antinausea medications that enabled costly chemotherapy previously contraindicated because of side effects. Of course, some patented treatments lower spending when they simply replace other, more costly treatments.43 Various newly patented drugs may produce all of these effects. Thus, the net impact of patents on total spending at each point in time remains unclear. Over time, however, there can be no doubt that patentinduced development of drugs and devices increases spending by expanding the range of feasible treatments and their effectiveness. Thus, the stakes in designing patent policy are huge. The progress of medical science and the growth of health spending depend in large measure on new products. How fast they are developed and how their benefits are distributed is profoundly influenced by the patent system. H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1271 O v e rv i e w n Research. A simple line of reasoning has led some to maintain that public support of basic biomedical research should be curtailed. The argument goes as follows: (1) technology has been the principal driver behind growing health care spending; (2) research indicates that some forms of high-technology medical care are overused; (3) growth of health care spending has become problematic in both private and public budgets. We believe that the correct conclusion from this line of reasoning is that the way Americans pay for health care should be changed but that if it is, the case for public support of biomedical research would be strengthened. The advance of medical technology is, indeed, the principal driver behind the admittedly problematic growth of health spending. It is, as we have noted, also the largest source of an even larger increase in benefits.7–12 At the same time, some medical services, including high-technology care, are overused or produced inefficiently. In recognition of these problems, Congress recently appropriated funds to renew and expand efforts to evaluate the comparative effectiveness of various medical interventions and to promote the use of IT. In addition, policymakers are now devoting considerable attention to the design of payment reforms to promote efficient and coordinated delivery of health care.44 Under the new payment systems, providers would no longer be paid for whatever services they rendered but instead would be paid for an episode of care, patients’ outcomes, or adherence to established protocols. The new payment systems would encourage cooperation among teams of caregivers instead of paying each provider without regard for cooperation and coordination. Such reforms are of particular importance in the handling of complex cases, which account for most health spending. With the curtailment of services that cost more than they are worth and improvements in the efficiency of care delivery, growth in total health spending would be reduced, but the net benefits of spending would increase. So would the expected net benefits from innovations. But incentives to invest in developing new drugs, devices, and procedures would diminish because returns to investments in innovation depend on the size of the anticipated market. Sales add to profits whether or not the sales are socially beneficial. Research is always a gamble that may or may not pay out, and reducing pay-offs makes gambles less attractive. Based on recent history, the average returns to medical innovations have been large despite a less-than-ideal payment system and a delivery system that is far from efficient. There is no reason to think that funders of basic research will be able in the future to predict reliably which investments will produce large benefits and which will not. There is, however, good reason to believe that those investing in development activities will be able to identify those advances from basic research that can be commercialized profitably. Consequently, there is a serious risk that payment reforms that improve the near-term efficiency of health care delivery would lower the longer-term rate of advance in medical knowledge. To offset this possibility, successful control of spending on low-benefit care should, in our view, be accompanied by increased public support of biomedical research to offset the 1272 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng reduced incentives for privately financed research. To be sure, returns to biomedical research, like those to all activities, must eventually diminish. Current scientific opportunities appear to be large as a result of major advances in molecular biology, materials sciences, and other fields relevant to medical advance. Nevertheless, inflation-adjusted public support for basic research has declined for about a decade. The proportion of all grant applications to the National Institutes of Health (NIH) that is funded has declined from just under one-third in 1999 to just over one-fifth in 2008.45 The economic stimulus bill (the American Recovery and Reinvestment Act) abruptly increased NIH funding by adding $10 billion, to be spent over fiscal years 2009 and 2010.46 Discussion And Implications For Reform The case that the United States spends more than is optimal on health care is overwhelming. Insurance can lead to excessive demand for care. Tax incentives encourage people to buy more insurance than they would if they paid full price for it. The fee-for-service reimbursement system creates economic incentives for oversupply and encourages fragmentation in the delivery of care. Evidence that too much of particular services is provided in some regions is persuasive. Opportunities to improve quality and lower costs remain unexploited. To lower spending without lowering net welfare, it is necessary to organize the delivery of care to promote efficient cooperation among the many providers and practitioners involved in delivering modern treatment, to conduct costly research over many years to identify which procedures are effective at reasonable cost, to develop protocols that enable providers to identify in advance patients in whom expected benefits of treatment are lower than costs, to design incentives that encourage providers to act on those protocols, and to educate patients on why such protocols should be sustained. Furthermore, if spending reductions are to prove beneficial over time, it is also necessary to provide research support to maintain the flow of beneficial innovations. Provider payment reform has a critical role to play in promoting efficient and coordinated delivery. Fee-for-service payment, especially when relative values are distorted, works against efficiency. Changes in the Medicare program have the potential for broad impact, since Medicaid programs and private insurers tend to follow Medicare methods. Reform would involve increasing both the degree to which current relative payments reflect relative costs of efficient production of services and the use of broader units of payments, such as per episode and per person, in either case likely blended with fee-for-service. Paul Ginsburg’s time was supported by the Robert Wood Johnson Foundation’s funding of the Center for Studying Health System Change. H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1273 O v e rv i e w NOTES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. OECD health data 2008. Version 12/10/2008. Available from http://fiordiliji.sourceoecd.org/v1=6937783/ c1=34/nw=1/rpsv/statistic/s37_about.htm?jnlissn=99991012 Ginsburg PB. High and rising health care costs: demystifying U.S. health care spending. Princeton (NJ): Robert Wood Johnson Foundation; 2008 Oct. Research Synthesis Report no. 16. Gerdtham UG, Jönsson B. International comparison of health expenditure: theory, data, and econometric analysis. In: Culyer AJ, Newhouse JP, editors. Handbook of health economics. Vol. 1A. New York (NY): Elsevier; 2000. p. 11–53. Farrell D, Kocher B, Laboissiere M, Parker S. Accounting for the cost of US health care: a new look at why Americans spend more. Washington (DC): McKinsey Global Institute; 2008 Dec. Fuchs VR. Who shall live? Health, economics, and social choice. New York (NY): Basic Books; 1974. Skinner J, Chandra A, Goodman D, Fisher ES. The elusive connection between health care spending and quality. Health Aff (Millwood). 2009;28(1):w119–23. Murphy KM, Topel RH. The value of health and longevity. J Politic Econ. 2006 Oct;114(5):871–904. Murphy KM, Topel RH, editors. Measuring the gains from medical research. Chicago (IL): University of Chicago Press; 2003. Nordhaus W. The health of nations: the contributions of improved health to living standards. In: Murphy KM, Topel RH, editors. Measuring the gains from medical research. Chicago (IL): University of Chicago Press; 2003. p. 9–40. Cutler DM, Rosen AB, Vijan S. The value of medical spending in the United States, 1960–2000. N Engl J Med. 2006 Aug 31;355(9):920–7. Hall RE, Jones CI. The value of life and the rise of health care spending. Quart J Econ. 2007;122(1):39–72. Garber AM, Skinner J. Is American health care uniquely inefficient? Cambridge (MA): National Bureau of Economic Research; 2008 Aug. Working Paper no. 14257. Chandra A, Staiger DO. Productivity spillovers in health care: evidence from the treatment of heart attacks. J Politic Econ. 2007 Feb;115(1):103–40. Ikegami N, Campbell JC. Health care reform in Japan: the virtues of muddling through. Health Aff (Millwood). 1999;18(3):56–75. Kogan R, Cox K, Horney JR. The long-term fiscal outlook is bleak: restoring fiscal sustainability will require major changes to programs, revenues, and the nation’s health care system. Washington (DC): Center on Budget and Policy Priorities; 2008 Dec. U.S. Congress. The long term budget outlook. Washington (DC): Congressional Budget Office; 2007 Dec. Board of Trustees, Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds. The 2008 annual report of the trustees of the federal hospital insurance and federal supplementary medical insurance trust funds. Washington (DC): U.S. Government Printing Office; 2008 Mar. Aaron HJ. Budget crisis, entitlement crisis, health care financing problem—which is it? Health Aff (Millwood). 2007;26(6):1622–33. Woolhandler S, Campbell T, Himmelstein DU. Costs of health care administration in the United States and Canada. N Engl J Med. 2003 Aug 21;349(8):768–75. Aaron HJ. The costs of health care administration in the United States and Canada—questionable answers to a questionable question. N Engl J Med. 2003 Aug 21;349(8):801–3. Orszag P. Behavioral economics: lessons from retirement research for health care and beyond. Proceedings of the annual meeting of the Retirement Research Consortium [Internet]. Washington (DC): Congressional Budget Office; 2008 Aug 7 [cited 2009 Jun 30]. Available from: http://crr.bc.edu/events/2008_ conference_agenda_and_papers.html Tu JV, Hannan EL, Anderson GM, Iron K, Wu K, Vranizan K, Popp AJ, et al. The fall and rise of carotid endarterectomy in the United States and Canada. N Engl J Med 1998;339(20):1441–7. Aaron HJ. Waste, we know you are out there. N Engl J Med. 2008 Oct 30;359(18):1865–7. Center for the Evaluation of Value and Risk in Health, Tufts Medical Center. Cost effectiveness analysis registry [Internet]. Boston (MA): Tufts Medical Center; [cited 2009 Jun 30]. Available from: https:// research.tufts-nemc.org/cear/default.aspx Dartmouth Institute for Health Policy and Clinical Practice. Dartmouth atlas of health care [Internet]. 1 274 September/October 2009 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. Health Sp endi ng Hanover (NH): Dartmouth College; c2009 [cited 2009 Jun 30]. Available from: http://www.dartmouth atlas.org 26. Boden WE, O’Rourke RA, Teo KK, Hartigan PM, Maron DJ, Kostuk WJ, et al. Optimal medical therapy with or without PCI for stable coronary disease. N Engl J Med. 2007: 26;356(15):1503–16. 27. Arad Y, Spadaro LA, Roth M, Guerci AD. Treatment of asymptomatic adults with elevated coronary calcium scores with atorvastatin, vitamin C, and vitamin E: the St. Francis Heart Study randomized clinical trial. J Am Coll Cardiol. 2005;46(1):166–72. 28. Arrow K. Uncertainty and the welfare economics of medical care. Amer Econ Rev. 1963 Dec;53:941–73. 29. Pauly M. The economics of moral hazard: a comment. Amer Econ Rev. 1968;58(1); part 3: 531–7. 30. Newhouse JP, Insurance Experiment Group. Free for all? Lessons from the RAND Health Insurance Experiment Group. Cambridge (MA): Harvard University Press; 1994. 31. Helms RB. Tax policy and the history of the health insurance industry. In: Aaron HJ, Burman LE, editors. Using taxes to reform health insurance. Washington (DC): Brookings Institution; 2008. p. 13–35. 32. Liebman J, Zeckhauser R. Simple humans, complex insurance, subtle subsidies. In: Aaron HJ, Burman LE, editors. Using taxes to reform health insurance. Washington (DC): Brookings Institution; 2008. p. 230– 62. 33. Berndt E. Pharmaceuticals in U.S. health care: determinants of quantity and price. J Econ Perspect. 2002 Aug;16(4):45–66. 34. Mongan JJ, Mechanic RE, Lee TH. Transforming U.S. health care: policy challenges affecting the integration and improvement of care. Washington (DC): Brookings Institution; 2006. 35. Ginsburg PB, Pham HH, McKenzie KL, Milstein A. Distorted payment system undermines business case for health quality and efficiency gains. Washington (DC): Center for Studying Health System Change; 2007 Jul. Issue Brief no. 112. 36. See, for example, Fuchs VR. Physician induced demand: a parable. J Health Econ. 1986;5(4):367. 37. Danzon PM. Liability for medical malpractice. In: Culyer AJ, Newhouse JP, editors. Handbook of health economics. 1st ed. Vol. 1. Amsterdam (Netherlands): Elsevier; 2000. p. 1309–1404. 38. Danzon PM. Liability for medical malpractice. In: Culyer AJ, Newhouse JP, editors. Handbook of health economics. 1st ed. Vol. 1. Amsterdam (Netherlands): Elsevier; 2000. p. 1339–1404. 39. Weiler PC, Hiatt HH, Newhouse JP, Johnson WG, Brennan T, Leape LL. A measure of malpractice: medical injury, malpractice litigation, and patient compensation. Cambridge (MA): Harvard University Press; 1993. 40. Clinton HR, Obama B. Making patient safety the centerpiece of medical liability reform. N Engl J Med. 2006 May 25;354(21):2205–8. 41. Calculations based on Centers for Medicare and Medicaid Services. National health expenditure projections 2008–2018. Table 1 [Internet]. Baltimore (MD): CMS; [cited 2009 Jun 30]. Available from: http:// www.cms.hhs.gov/NationalHealthExpendData/downloads/proj2008.pdf 42. Danzon PM, Furukawa MF. International prices and availability of pharmaceuticals in 2005. Health Aff (Millwood). 2008;27(1):221–33. 43. Lichtenberg FR. Effects of new drugs on overall health spending: Frank Lichtenberg responds. Health Aff (Millwood). 2007;26(3):887–90. 44. Medicare Payment Assessment Commission. Report to the Congress: reforming the delivery system [Internet]. Washington (DC): MedPAC; 2008 [cited 2009 Jun 30]. Available from: http://www.medpac .gov/documents/Jun08_EntireReport.pdf 45. National Institutes of Health. Research Portfolio Online Reporting Tool (RePORT) [Internet]. Bethesda (MD): NIH; [cited 2009 Jun 30]. Available from: http://report.nih.gov/reports.aspx?section=SuccessRates &title=Success%20Rate 46. Steinbrook R. Health care and the American Recovery and Reinvestment Act. N Engl J Med. 2009:360(11): 1057–60. H E A L T H A F F A I R S ~ Vo l u m e 2 8 , N u m b e r 5 Downloaded from HealthAffairs.org on January 21, 2020. Copyright Project HOPE—The People-to-People Health Foundation, Inc. For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org. 1275 Health Insurance and the Demand for Medical Care: Evidence from a Randomized ... Manning, Willard G.; Newhouse, Joseph P.; Duan, Naihua; Keeler, Emmett B.; Le... The American Economic Review; Jun 1987; 77, 3; ABI/INFORM Global pg. 251 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. American Economic Review 2013, 103(7): 2643–2682 http://dx.doi.org/10.1257/aer.103.7.2643 Adverse Selection and Inertia in Health Insurance Markets: When Nudging Hurts† By Benjamin R. Handel* This paper investigates consumer inertia in health insurance markets, where adverse selection is a potential concern. We leverage a major change to insurance provision that occurred at a large firm to identify substantial inertia, and develop and estimate a choice model that also quantifies risk preferences and ex ante health risk. We use these estimates to study the impact of policies that nudge consumers toward better decisions by reducing inertia. When aggregated, these improved individual-level choices substantially exacerbate adverse selection in our setting, leading to an overall reduction in welfare that doubles the existing welfare loss from adverse selection. (JEL D82, G22, I13) A number of potential impediments stand in the way of efficient health insurance markets. The most noted of these is adverse selection, first studied by Akerlof (1970) and Rothschild and Stiglitz (1976). In insurance markets, prices reflect the expected risk (costs) of the insured pool. Whether the reason is price regulation or private information, when insurers cannot price all risk characteristics riskier consumers choose more comprehensive health plans. This causes the equilibrium prices of these plans to rise and healthier enrollees to select less comprehensive coverage than they would otherwise prefer. * Department of Economics, University of California at Berkeley, 530 Evans Hall #3880, Berkeley, CA 94720 (e-mail: handel@berkeley.edu). I thank my dissertation committee chairs Igal Hendel and Michael Whinston for their guidance on this project. My third committee member, David Dranove, provided invaluable advice throughout the research process. I thank three anonymous referees for their advice and substantial effort in helping me improve the paper throughout the review process. Also, I thank conference discussants Gautam Gowrisankaran, Jonathan Gruber, Amanda Kowalski, and Robert Town for their advice and effort. This work has benefited particularly from the extensive comments of Zarek Brot-Goldberg, Leemore Dafny, Stefano DellaVigna, J. P. Dube, Liran Einav, Amy Finkelstein, Ron Goettler, Kate Ho, Mitch Hoffman, Kei Kawai, Jon Kolstad, Jonathan Levin, Neale Mahoney, Kanishka Misra, Aviv Nevo, Mallesh Pai, Ariel Pakes, Rob Porter, James Roberts, Bill Rogerson, and Glen Weyl. I have received invaluable advice from numerous others including my colleagues at Berkeley and Northwestern as well as seminar participants at the 2011 AEA Meetings, ASHE-Cornell, the Bates White Antitrust Conference, Berkeley, Booth School of Business, CalTech, Columbia, the Cowles Structural Microeconomics Conference, EUI, Haas School of Business, Harvard, Harvard Business School, Harvard Kennedy School, the HEC Montreal Health-IO Conference, the Milton Friedman Institute Health Economics Conference, M.I.T., Microsoft Research, Northwestern, Olin School of Business, Princeton, RAND, Sloan School of Management, Stanford Business School, Toronto, Toulouse School of Economics, UBC, UCSD, UC-Davis, University of Chicago, University of Michigan, University of Warwick, Yale, and Yale SOM. I gratefully acknowledge funding support from the Center for the Study of Industrial Organization (CSIO) at Northwestern and the Robert Wood Johnson Foundation. All remaining errors are my own. I obtained approval for the acquisition and use of the data for this paper from the Northwestern IRB while a student at Northwestern. The author has no financial or other material interests related to this research to disclose. † Go to http://dx.doi.org/10.1257/aer.103.7.2643 to visit the article page for additional materials and author disclosure statement(s). 2643 2644 THE AMERICAN ECONOMIC REVIEW december 2013 A second less studied, but potentially important, impediment is poor health plan choice by consumers. A collection of research summarized by Thaler and Sunstein (2008) presents strong evidence that consumer decisions are heavily influenced by context and can systematically depart from those that would be made in a rational frictionless environment. These decision-making issues may be magnified when the costs and benefits of each option are difficult to evaluate, as in the market for health insurance. In the recently passed Affordable Care Act (ACA), policymakers emphasized clear and simple standardized insurance benefit descriptions as one way to improve consumer choices from plan menus offered through proposed exchanges. If consumers do not have the information or abilities to adequately choose an insurance plan, or have high tangible search or switching costs, there can be an immediate efficiency loss from consumers not maximizing their individual well-being as well as a long term efficiency loss from not transmitting the appropriate price signals to the competitive marketplace. In this work we empirically investigate how one source of choice inadequacy, inertia, interacts with adverse selection in the context of an employer-sponsored insurance setting typical of the US health care system.1 In health insurance markets, this interaction matters because choice adequacy impacts plan enrollment, which in turn determines average costs and subsequent premiums. Thus, if there are substantial barriers to decision-making this can have a large impact on the extent of adverse selection and, consequently, consumer welfare. Policies designed to improve consumer choice will have a theoretically ambiguous welfare effect as the impact of better decision making conditional on prices could be offset by adverse selection, if it is exacerbated. This stands in contrast to most previous work on choice inadequacy where policies designed to improve consumer choices can only have positive welfare impacts. We study individual-level health plan choice and health claims data for the employees of a large firm and their dependents. The data contain a major change to insurance provision that we leverage to identify inertia separately from persistent consumer preference heterogeneity. The firm implemented this change to their employee insurance program in the middle of the six years of data we observe. The firm significantly altered their menu of five health plan offerings, forced employees out of the health plans they had been enrolled in, and required them to actively choose a plan from the new menu, with no stated default option. In subsequent years, the insurance plan options remained the same but consumers had their previously chosen plan as a default option, implying they would continue to be enrolled in that plan if they took no action. This was despite the fact that employee premiums changed markedly over time such that many would have benefited from switching their plan. When combined with other features of the data, our ability to observe the same consumers in clearly active and clearly passive choice environments over time allows us to cleanly identify inertia. Since the plans that we study have the same network of providers and cover the same 1 In 2009, 55.8 percent of all individuals in the United States (169 million people) received insurance through their employer or the employer of a family member (DeNavas-Walt, Proctor, and Smith 2010). The amount of money at stake in this setting is large: in 2010 the average total premium (employer plus employee contribution) for an employer provided insurance plan was $5,049 for single coverage and $13,770 for family coverage (Kaiser Family Foundation 2010a). VOL. 103 NO. 7 handel: adverse selection and inertia 2645 medical services, the inertia we measure does not come from an unwillingness to switch medical providers, which is an important factor in many settings. We present descriptive tests that suggest the presence of substantial inertia. Our first test for inertia studies the behavior of new employees at the firm. As plan prices and the choice environment change over time, incoming cohorts of new employees make active choices that reflect the updated setting while prior cohorts of new employees make markedly different choices that reflect the past choice setup, though they are similar on all other dimensions. A second test studies specific cases that arise in our environment where certain groups of consumers have one of their health plan options become completely dominated by another due to price changes over time. The majority of consumers who face this scenario continue to choose a plan once it becomes dominated, despite the fact that all of them should switch in a frictionless market. Additionally, we present a test for adverse selection revealing that higher health risk employees choose more comprehensive coverage. While these tests show that inertia and adverse selection are important in our ­environment, to precisely measure these effects and understand the impact of ­counterfactual policies we develop a structural choice model that jointly quantifies inertia, risk preferences, and ex ante health risk. In the model, consumers make choices that maximize their expected utilities over all plan options conditional on their risk tastes and health risk distributions. In the forced active choice period consumers have no inertia (by construction), while in periods that have an incumbent plan option inertia reduces the utility of alternative options relative to the status quo option. While there are several potential micro-foundations for inertia, we model inertia as the implied monetary cost of choice persistence, similar in structural interpretation to a tangible switching cost.2 We allow for heterogeneity in both inertia and risk preferences so that we have the richest possible understanding of how consumers select plans. To model health risk perceived by employees at the time of plan choice, we develop an out-of-pocket expense model that leverages sophisticated predictive software developed at Johns Hopkins Medical School. The model uses detailed past diagnostic and cost information to generate individual-level and plan-specific expense risk projections that represent ex ante uncertainty in the choice framework. Our choice model estimates reveal large inertia with some meaningful heterogeneity, modeled as a function of observable family characteristics. In our primary specification, inertia causes an average employee to forgo $2,032 annually, while the population standard deviation is $446 (an average employee’s family spends $4,500 each year). An employee covering at least one dependent forgoes, on average, $751 more than a single employee while an employee that enrolls in a flexible spending account (FSA), an account that requires active yearly participation, forgoes $551 less than one who does not. Our risk preference estimates reveal that consumers have a meaningful degree of risk aversion, suggesting that there are, on average, substantial benefits from incremental insurance. We present a variety of 2 We discuss potential sources of inertia and their implications for our framework further in Section III, in the context of the choice model, and in online Appendix D. Search costs, switching costs, and psychological costs are examples of potential micro-foundations for inertia, each of which could imply a different underlying choice model. In this work, we do not attempt to distinguish between distinct underlying sources of inertia. 2646 THE AMERICAN ECONOMIC REVIEW december 2013 robustness analyses to demonstrate that our parameter estimates are quite stable with respect to some of the underlying assumptions in our primary specification. We use these estimates to study a counterfactual policy intervention that reduces inertia from our baseline estimates. This counterfactual analysis is intended to apply broadly to any proposed policies that have the potential to decrease inertia: targeted information provision, premium and benefits change alerts, and standardized and simplified insurance plan benefit descriptions are three oft-discussed policies. We take for granted that there are a range of potential policies that differentially reduce inertia, and that these policies reduce inertia through the mechanism assumed in our primary empirical specification.3 We examine a range of policy interventions spanning the case where the extent of inertia is unchanged to the case where it is completely eliminated. In order to assess the impact of reduced inertia, it is necessary to model the supply-side of the insurance market. To this end, we construct an insurance pricing model that closely follows the way premiums were determined in the firm we study. In our framework, plan premiums equal the average costs of enrollees from the prior period plus an administrative fee, conditional on the number of dependents covered. The firm provides employees with a flat subsidy toward these premiums, implying that consumers pay the full marginal cost of more comprehensive insurance. This pricing environment is very similar to that studied in prior work on insurance markets by, e.g., Cutler and Reber (1998) and Einav, Finkelstein, and Cullen (2010). It also closely resembles the competitive environment of the insurance exchanges recently proposed in the ACA, though there are some specific differences we highlight. In the naïve case where plan prices do not change as a result of the different enrollment patterns caused by the intervention, a three-quarter reduction in inertia substantially improves consumer choices over time. This reduction leads to a $105 mean per person per year welfare increase, which equals 5.2 percent of the mean employee premium paid. In the primary policy analysis, where insurance prices endogenously respond to different enrollment and cost patterns, the results are quite different. The same policy that reduces inertia by three-quarters still improves consumer choices conditional on prices, but now also exacerbates adverse selection, leading to a 7.7 percent reduction in welfare.4 In this more fluid marketplace, consumers who are healthy and value comprehensive insurance can no longer reasonably purchase it because of the high relative premiums caused by acute sorting. This intervention essentially doubles the existing 8.2 percent welfare loss from adverse selection in our observed environment, a figure that much of the literature focuses on. We also find that welfare is decreasing as the intervention to reduce inertia becomes more effective. There are substantial distributional consequences resulting from the reduction in inertia, in addition to the overall efficiency loss. 3 In order to determine the impact that specific policies will have in reducing inertia, it is important to distinguish between potential underlying mechanisms for inertia. Here, we focus on the overall magnitude of inertia and its interaction with adverse selection and assume one specific inertial mechanism. We argue later that, given the source of identification, the counterfactual analysis would yield similar results with different underlying inertial mechanisms. 4 Our welfare analysis accounts for the different potential underlying sources of inertia by considering a s­ pectrum of cases ranging from the one where switching plans represents a true social cost (e.g., tangible switching or search costs) to the case where switching only matters for the resulting choices and is not a cost in and of itself (e.g., unawareness/inattention). The welfare impact is negative across this spectrum for almost all policy interventions. VOL. 103 NO. 7 handel: adverse selection and inertia 2647 It is important to note that the negative welfare impact from reduced inertia that we find is specific to our setting on multiple dimensions. First, we study a specific population with specific preferences and health risk profiles: the direction of the welfare impact could be reversed with a different population in the same market environment. Second, the market environment that we study is specific: the direction of the welfare impact could be reversed with the same population in a different market environment. Nevertheless, the analysis clearly illustrates that the interaction between adverse selection and inertia can have substantial, and potentially surprising, welfare implications. This paper contributes to several distinct literatures. The clean identification of inertia that we obtain from the plan re-design and forced active re-enrollment resolves a primary issue in the empirical literature that seeks to quantify the implicit monetary value of inertia and related phenomena. Farrell and Klemperer (2007) survey related work on switching costs and discuss how the inability of researchers to observe active or initial choices within a micro-level panel dataset confounds their ability to separately identify switching costs from persistent unobserved preference heterogeneity. Shum (2004); Crawford, Tosini, and Waehrer (2011); and Goettler and Clay (2011) are recent studies in this vein that study switching costs in the context of breakfast cereals, fixed-line telephone plans, and grocery delivery markets respectively. Dube et al. (2008) and Dube, Hitsch, and Rossi (2010) are examples of related work in the marketing literature on brand loyalty and state dependence. There is also relevant work that studies the effects of inertia without explicitly quantifying its value (see, e.g., Strombom, Buchmueller, and Feldstein 2002; and Ericson 2012 in health insurance and Madrian and Shea 2001 in 401(k) plan choice). Our work differs from this latter literature on several dimensions, including that (i) we explicitly quantify the value of inertia and other micro-foundations and (ii) we use those estimates to study the interaction between inertia and adverse selection. It is important to note that, while sometimes using different terminology, these prior papers study similar factors leading to choice persistence beyond stable innate preferences. As in this paper, these prior papers do not distinguish between distinct sources of inertia. This analysis also builds on the prior work that studies the existence and consequences of adverse selection in health insurance markets. Our insurance choice model relates to the approach of Cardon and Hendel (2001), which is also similar to the approaches used in Carlin and Town (2009), Bundorf, Levin, and Mahoney (2012), and Einav et al. (2013). These papers model selection as a function of expected health risk and study the welfare loss from adverse selection in their observed settings relative to the first-best. Our work adds to this literature by quantifying inertia and investigating its interaction with adverse selection. With different underlying empirical frameworks, Cutler and Reber (1998) and Einav, Finkelstein, and Cullen (2010) also study the welfare consequences of adverse selection in the context of large self-insured employers. Another relevant strand of work studies the impact of preference dimensions separate from risk on adverse (or advantageous) selection. Cutler, Finkelstein, and McGarry (2008); Cutler, Lincoln, and Zeckhauser (2010); Fang, Keane, and Silverman (2008); and Einav et al. (2013) study alternative dimensions of selection in health insurance markets (e.g., risk preferences and moral hazard) while Cohen and Einav (2007) and Einav, Finkelstein, 2648 THE AMERICAN ECONOMIC REVIEW december 2013 and Schrimpf (2010) study such dimensions in auto insurance and annuity markets, respectively. For a more in depth discussion of these literatures see the recent survey by Einav, Finkelstein, and Levin (2010). The rest of the paper proceeds as follows. Section I describes the data with an emphasis on how the health insurance choice environment evolves at the firm over time. Section II presents simple descriptive tests that show the presence of both inertia and adverse selection. Section III presents our empirical framework while Section IV presents the structural estimates from this model. Section V presents a model of insurance pricing, describes our welfare framework, and investigates the impact of counterfactual policies that reduce inertia. Section VI concludes. I. Data and Environment We study the health insurance choices and medical utilization for the employees at a large US based firm, and their dependents, over the time period from 2004 to 2009. In a year during this period that we denote t?0?? (to protect the identity of the firm) the firm changed the menu of health plans it offered to employees and introduced an entirely new set of PPO plan options.5 At the time of this change, the firm forced all employees to leave their prior plan and actively re-enroll in one of five options from the new menu, with no stated default option. The firm made a substantial effort to ensure that employees made active choices at ?t0??by continuously contacting them via physical mail and e-mail to both communicate information about the new insurance program and remind them to make a choice.6 In the years prior to and following the active choice year ?t?0??, employees were allowed to default into their previously chosen plan option without taking any action, despite the fact that in several cases plan prices changed significantly. This variation in the structure of the default option over time, together with the plan menu change, is a feature of the dataset that makes it especially well suited to study inertia because, for each longerterm employee, we observe at least one choice where inertia could be present and one choice where it is not. These proprietary panel data include the health insurance options available in each year, employee plan choices, and detailed, claim-level, employee and dependent medical expenditure and utilization information.7 We use this detailed medical information together with medical risk prediction software developed at Johns Hopkins Medical School to develop individual-level measures of projected future medical utilization at each point in time. These measures are generated using past diagnostic, expense, and demographic information and allow us to precisely gauge medical expenditure risk at the time of plan choice in the context of our cost model.8 5 This change had the two stated goals of (i) encouraging employees to choose new, higher out-of-pocket spending plans to help control total medical spending and (ii) providing employees with a broader plan choice set. 6 Ultimately, 99.4 percent of employees ended up making an active choice. Although they were not told about a default option ahead of time, the 0.6 percent employees that did not actively elect a plan were all enrolled in one of the new plan options, PP?O500? ?. 7 We observe detailed medical data for all employees and dependents enrolled in one of several PPO options, the set of available plans our analysis focuses on. These data include detailed claim-level diagnostic information (e.g., ICD-9 and NDC codes), provider information, and payment information (e.g., deductible paid, plan paid). 8 The Johns Hopkins ACG (Adjusted Clinical Groups) Case-Mix System is widely used in the health care sector and was specifically designed to incorporate individual-level diagnostic claims data to predict future medical expenditures in a sophisticated manner (e.g., accounting for chronic conditions). VOL. 103 NO. 7 handel: adverse selection and inertia 2649 Additionally, we observe a rich set of employee demographics including job characteristics, age, gender, income, and job tenure, along with the age, gender, and type of each dependent. Together with data on other relevant choices (e.g., flexible spending account (FSA) contributions, dental insurance) we use these characteristics to study heterogeneity in inertia and risk preferences. Sample Composition and Demographics.—The firm we study employs approximately 9,000 people per year. The first column of Table 1 describes the demographic profile of the 11,253 employees who work at the firm for some stretch within 2004 –2009. These employees cover 9,710 dependents, implying a total of 20,963 covered lives. 46.7 percent of the employees are male and the mean employee age is 40.1 (median of 37). We observe income grouped into five tiers, the first four of which are approximately $40,000 increments, increasing from zero, with the fifth for employees that earn more than $176,000. Almost 40?­percent of employees have income in tier 2, between $41,000 and $72,000, with 34 percent less than $41,000 and the remaining 26 percent in the three income tiers greater than $72,000. Fifty-eight percent of employees cover only themselves with health insurance, with the other 42 percent covering a spouse and/or dependent(s). Twenty-three percent of the employees are managers, 48 percent are white-collar employees who are not managers, and the remaining 29 percent are blue-collar employees. Thirteen?­percent of the employees are categorized as “quantitatively sophisticated” managers.9 Finally, the table presents information on the mean and median characteristics of the zip codes the employees live in. We construct our final sample to leverage the features of the data that allow us to identify inertia. Moving from the full data, we restrict the final sample to employees and dependents who (i) are enrolled in a health plan for all years from t??−1? to?t?1? and (ii) are enrolled in a PPO option in each of those years (this excludes the employees who enroll in either of two HMO options).10 The second column in Table 1 describes the sample of employees who ever enroll in a PPO option at the firm (N = 5,667), while the third column describes the final sample (N = 2,023). Comparing column 2 to column 1, it is evident that the restriction to PPO options engenders minimal selection based on the rich set of demographics we observe. Comparing both of these columns to column three reveals that the additional restriction that employees be enrolled for three consecutive years does lead to some sample selection: employees in the final sample are slightly older, slightly richer, and more likely to cover additional family members than the overall PPO population. Note that the multi-year enrollment restriction primarily excludes employees who enter or exit the firm during this period, rather than those who switch to an HMO option or waive coverage. There are costs and benefits of these two restrictions. The restriction to PPO plans is advantageous because we observe detailed medical claims data only for enrollees in these plans and these plans are only differentiated by financial characteristics, implying we don’t have to consider heterogeneity in preferences over provider network when modeling choice between them. A potential cost is that this restriction 9 These are managers associated with specific groups where the work is highly quantitative in nature. We denote all years in reference to ?t0??, such that, e.g., year ?t−1? ?occurred just before t?0??and year t?1??just after. 10 2650 december 2013 THE AMERICAN ECONOMIC REVIEW Table 1—Descriptive Statistics Sample demographics N–Employee only N–All family members Mean employee age (median) Gender (male) percent Income ( percent) Tier 1 (< $41K) Tier 2 ($41K–$72K) Tier 3 ($72K–$124K) Tier 4 ($124K–$176K) Tier 5 (> $176K) Family size ( percent) 1 2 3 4+ Staff grouping ( percent) Manager (percent) White-collar (percent) Blue-collar (percent) Additional demographics Quantitative manager (percent) Job tenure mean years (median) Zip code population mean (median) Zip code income mean (median) Zip code house value mean (median) All employees PPO ever Final sample 11,253 20,963 40.1 (37) 46.7 5,667 10,713 40.0 (37) 46.3 2,023 4,544 42.3 (44) 46.7 33.9 39.5 17.9 5.2 3.5 31.9 39.7 18.6 5.4 4.4 19.0 40.5 25.0 7.8 7.7 58.0 16.9 11.0 14.1 56.1 18.8 11.0 14.1 41.3 22.3 14.1 22.3 23.2 47.9 28.9 25.1 47.5 27.3 37.5 41.3 21.1 12.8 7.2 (4) 42,925 (42,005) $56,070 ($55,659) $226,886 ($204,500) 13.3 7.1 (3) 43,319 (42,005) $56,322 ($55,659) $230,083 ($209,400) 20.7 10.1 (6) 41,040 (40,175) $60,948 ($57,393) $245,380 ($213,300) Notes: This table presents summary demographic statistics for the population we study. The first column describes demographics for the entire sample, whether or not they ever enroll in insurance with the firm. The second column summarizes these variables for the sample of individuals who ever enroll in a PPO option, the choices we focus on in the empirical analysis. The third column describes our final estimation sample, which includes those employees who (i) are enrolled in PP?O−1? ?at ?t?−1?and (ii) remain enrolled in any plan at the firm through at least ?t1? ?. Comparing the columns shows little selection on demographics into PPO options and some selection based on family size into the final estimation sample. may bias the choice model by restricting the set of options. In the upcoming descriptive analysis of plan choices we show clear evidence that the nest of PPO options and nest of HMO options are quite horizontally differentiated from one another, implying a limited within-sample bias from excluding HMO choices. The restriction that employees enroll in a plan in every year from t??−1?to ?t1?? has the benefit that, for each individual in the final sample, we observe a past year of medical data for each choice spanning ?t0??to ?t2???. This allows us to model health risk at the time of each choice from an ex ante perspective, permitting a more precise characterization of out-of-pocket expense risk and the choice model parameters. This restriction has two costs: (i) it reduces the sample size and (ii) it excludes new VOL. 103 NO. 7 handel: adverse selection and inertia 2651 employees from t??0?to t??2??, who, as the upcoming preliminary analysis section reveals, can provide an additional source of identification for inertia. Ultimately, since the identification within the final sample for inertia is quite strong because of the plan menu change and linked active decision, we feel that having a more precise model is worth the costs of this restriction.11 Health Insurance Choices.—From 2004 to t?−1? ?the firm offered five total health plan options composed of four HMO plans (restricted provider network, greater cost control) and one PPO plan (broader network, less cost control). Each of these five plans had a different network of providers, different contracts with providers, and different premiums and cost-sharing formulas for enrollees. From t? ?0?on, the new plan menu contained two of the four incumbent HMO plans and three new PPO plans.12 This plan structure remained intact through the end of the data in 2009. After the menu change, the HMOs still had different provider networks and cost sharing rules both relative to each other and to the set of new PPOs. However, the three new PPO plans introduced at ?t?0?had exactly the same network of providers, the same contractual treatment of providers, and cover the same medical services. The PPO plans are only differentiated from one another (and from the previously offered PPO) by premiums and cost sharing characteristics (e.g., deductible, coinsurance, and out-of-pocket maximums) that determine the mapping from total medical expenditures to employee out-of-pocket expenditures. Throughout the period, all PPO options that the firm offers are self-insured plans where the firm fills the primary role of the insurer and is at risk for incurred claims. We denote the HMO plans available throughout the entire period as HM?O?1? and HM?O?2?, and those offered only prior to ?t0?? as HM?O3?? and HM?O4??. We denote the PPO option from before the menu change as PP?O?−1??, while we denote each of the PPO options after the menu change by their respective individual-level deductibles: PP?O2?50?? , PP?O5?00?? , and PP?O1?200?.13 PP?O?1200?is paired with a health savings account (HSA) option that allows consumers to deposit tax-free dollars to be used later to pay medical expenditures.14 Table A-2 in online Appendix E presents the detailed characteristics of the PPO plans offered at the firm over time. After the deductible is paid, PP?O?250?has a coinsurance rate of 10 percent while the other two plans have rates of 20 percent, implying they have double the marginal price of post-deductible claims. Out-of-pocket maximums indicate the maximum amount of medical expenditures that an enrollee can pay post-premium in a given plan. These are larger the less comprehensive the plan is and vary with income tier. Finally, both PP?O?250? and PP?O?500?have the same flat-fee co-payment structures for pharmaceuticals and physician office visits, while 11 We could include new employees after t? ?−1?using a less precise cost framework based on, e.g., age, gender, and future claims, similar to what is done in the literature when detailed claims data are not available. 12 An employee who chose a PPO plan at t? ?0???, by construction, actively chose a new plan. An employee who chose an incumbent HMO prior to t??0?was forced out of that plan and prompted to make an active choice from the new menu, though their old plan remained available. Since we only study PPO plans after t??0???, the incumbent aspect of the HMO plans does not impact our analysis. 13 The deductibles are indicative of how comprehensive (level of insurance) each plan is: for example, PP?O?250?provides the most insurance and has the highest premium. 14 This may lead to some horizontal differentiation for PP?O?1200?relative to the other two, which we account for in the choice model. This kind of plan is known as a “consumer driven health plan” (CDHP). Employees who signed up for this plan for the first time were given up to a $1,200 HSA match from the firm, which our analysis accounts for. 2652 THE AMERICAN ECONOMIC REVIEW december 2013 in PP?O1200? ?these apply to the deductible and coinsurance.15 Though we model these characteristics at a high-level of detail, our cost model necessarily makes some simplifying assumptions that we discuss and validate in online Appendix A.? ?graphically, to illustrate the Figure 1, panel A compares plans PP?O?250? and PP?O500 relationship between health plan financial characteristics, total medical expenses, and employee expenses. The figure applies specifically to year t??0? (premiums differ by year) and to low-income families, though it looks similar in structure for other coverage tiers and income levels. It completely represents the in-network differences between these two plans, since they are identical on the dimensions excluded from this chart, such as co-payments for pharmaceuticals and office visits. For this figure, and the rest of our analysis, we assume that (i) premiums are in pre-tax dollars and (ii) medical expenses are in post-tax dollars.16 After the employee premium, as total expenditures increase each employee pays the plan deductible, then the flat coinsurance rate, and finally has zero marginal cost after reaching the out-of-pocket maximum.17 As expected, the chart reveals that, ex post, healthy employees should have chosen PP?O?500?and sick employees PP?O2?50?? . Table A-3 in online Appendix E presents details on the pattern of employee choices over time before and after the menu change, which we summarize here. In ?t−? 1??, 39 percent of employees enroll in PP?O?−1??, 47 percent enroll in one of the four HMO options, and 14 percent waive coverage. At t? ?0??, 46 percent of employees choose one of the three new PPO options, with 25 percent choosing PP?O?250?? . 37?­percent choose either of the two remaining HMO plans while 16 percent waive coverage. Table A-3 also presents clear evidence that the nest of PPO options and nest of HMO options are quite horizontally differentiated from one another by examining consumer health plan transitions over time. An individual who switches plans from a PPO option is much more likely to choose another PPO option than to choose an? 1?who also enroll HMO option. Of the 2,757 employees enrolled in PP?O?−1?in year ?t− in any plan at t? 0???, only 85 (3 percent) choose an HMO option at ?t0???. In reverse, despite the expansion of PPO options and reduction of HMO options, only 15 percent of employees who chose an HMO option in t?−? 1?, and choose any plan at t?0???, switch to a PPO option. This suggests that restricting the set of choices to PPO options should not lead to biased parameters within that population. Each plan offered by the firm has a distinct total premium and employee premium contribution in each year. The total premium is the full cost of insurance while the employee premium contribution is the amount the employee actually pays after 15 These characteristics are for in-network purchases: the plans also have out-of-network payment policies, which we do not present or model. The plans have reasonably similar out-of-network payment characteristics (including out-of-pocket maximums). Only 2 percent of realized total expenditures are out-of-network. 16 In reality, medical expenses could also be in pre-tax dollars since individuals can pay medical expenses with pre-tax FSA or HSA contributions. In our data, 25 percent of the population enrolls in these accounts, which fund an even lower percentage of overall employee expenses. We convert premiums into pre-tax dollars by multiplying them by an income and family status contingent combined state and federal marginal tax rate using the NBER TAXSIM data. We may understate marginal tax rates of employees with high-earning spouses, since we don’t observe spousal income. 17 Each family member technically has his or her own deductible and out-of-pocket maximum. Families with 3+ members have aggregate deductible and out-of-pocket caps that bind if multiple family members reach their individual limits. While we explicitly take this structure into account in our cost model, Figure 1 assumes proportional allocation of expenses across family members. VOL. 103 NO. 7 2653 handel: adverse selection and inertia Panel A. PPO health insurance plan characteristics, t0 low-income family 8,000 PPO500 out-of-pocket maximum Total employee expenses 7,000 6,000 PPO250 out-of-pocket maximum 5,000 Coinsurance 4,000 PPO250 PPO500 Deductible 3,000 Premium 2,000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 50 ,00 ,50 ,00 7,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 7,00 ,50 ,00 3 4 6 9 10 12 13 15 16 18 19 21 22 24 25 2 28 30 1, t0 in-network total medical expenses* Panel B. PPO health insurance plan characteristics, t1 low-income family PPO250 out-of-pocket maximum 8,000 Coinsurance Total employee expenses 7,000 6,000 5,000 4,000 PPO500 out-of-pocket maximum PPO250 3,000 PPO500 2,000 Deductible Premium 1,000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 50 ,00 ,50 ,00 7,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 ,00 ,50 7,00 ,50 ,00 3 4 6 9 10 12 13 15 16 18 19 21 22 24 25 2 28 30 1, t1 in-network total medical expenses* Figure 1. Financial Characteristics of PPO250 and PPO500?Notes: This figure describes the relationship between total medical expenses (plan plus employee) and employee out-of-pocket expenses in years t0 and t1 for PPO250 and PPO500. This mapping depends on employee premium, deductible, coinsurance, and out of pocket maximum. This chart applies to low-income families (premiums vary by number of dependents covered and income tier, so there are similar charts for all 20 combinations of these two variables). Premiums are treated as pre-tax expenditures while medical expenses are treated as post-tax. Panel B presents the analogous chart for time t1 when premiums changed significantly, which can be seen by the change in the vertical intercepts. At time t0 healthier employees were better off in PPO500 and sicker employees were better off in PPO250. At time t1 all employees that this figure applies to should choose PPO500 regardless of their total claim levels, i.e., PPO250 is dominated by PPO500. Despite this, many employees who chose PPO250 in t0 continue to do so at t1, indicative of high inertia. *Total medical expenses equals plan paid plus employee paid. Ninety-six percent of all expenses are in network. 2654 december 2013 THE AMERICAN ECONOMIC REVIEW Panel A. PPO employee premiums, individual tier 4,000 Premium ($) 3,500 t0 +$420 3,000 t1 t2 2,500 –$288 2,000 –$324 1,500 1,000 500 0 1 2 3 4 PPO250 5 1 Income tier 2 3 4 PPO500 5 1 Income tier 2 3 4 5 PPO1200 Panel B. PPO employee premiums, family tier 12,000 Premium ($) t0 t1 t2 +$1,020 10,000 8,000 –$1,668 –$1,560 6,000 4,000 2,000 0 1 2 3 4 PPO250 5 1 Income tier 2 3 4 PPO500 5 1 Income tier 2 3 4 5 PPO1200 Figure 2. Evolution of Health Plan Premiums Notes: This figure describes the evolution of employee premium contributions at the firm over time between years t0 and t2. Employee premium contributions depend both on the number of dependents covered and the employee income tier. Panel A describes premiums for single employees and panel B relates to families (employee + spouse + dependent(s)). The figure illustrates the large relative employee premium contribution changes between t0 and t1 across tiers. receiving a subsidy from the firm.18 Total premiums are conditioned on being in one of four coverage tiers.19 The firm conditions PPO subsidies on an employee’s income tier, presumably because of equity considerations.20 Figure 2 illustrates employee premium contributions in years t? ?0?and t? 1??for the single and family (spouse plus children) coverage tiers. There is a noticeable decrease in premiums for PP?O5?00?from ?t?0?to ?t?1?coupled with an increase in the premium for PP?O?250??.