Moral Hazard in Health Insurance 9780231538688

Citation preview

MORAL HAZARD IN HEALTH INSURANCE

KENNETH J. ARROW LECTURE SERIES

KENNETH J. ARROW LECTURE SERIES

Kenneth J. Arrow’s work has shaped the course of economics for the past sixty years so deeply that, in a sense, every modern economist is his student. His ideas, style of research, and breadth of vision have been a model for generations of the boldest, most creative, and most innovative economists. His work has yielded such seminal theorems as general equilibrium, social choice, and endogenous growth, proving that simple ideas have profound effects. The Kenneth J. Arrow Lecture Series, supported by the Committee on Global Thought and the Program for Economic Research, highlights economists, from Nobel laureates to groundbreaking younger scholars, whose work builds on Arrow’s scholarship as well as his innovative spirit. The books in the series are an expansion of the lectures that are held in Arrow’s honor at Columbia University. Creating a Learning Society: A New Approach to Growth, Development, and Social Progress, Joseph E. Stiglitz and Bruce C. Greenwald The Arrow Impossibility Theorem, Eric Maskin and Amartya Sen Speculation, Trading, and Bubbles, José A. Scheinkman

MORAL HAZARD IN HEALTH INSURANCE AMY FINKELSTEIN WITH KENNETH J. ARROW, JONATHAN GRUBER, JOSEPH P. NEWHOUSE, AND JOSEPH E. STIGLITZ

COLUMBIA UNIVERSITY PRESS | NEW YORK

Columbia University Press Publishers Since 1893 New York Chichester, West Sussex cup.columbia.edu Copyright © 2015 Columbia University Press All rights reserved Library of Congress Cataloging-in-Publication Data Finkelstein, Amy. Moral hazard in health insurance / Amy Finkelstein. pages cm. — (Kenneth J. Arrow lecture series) Includes bibliographical references and index. ISBN 978-0-231-16380-4 (cloth : alk. paper) — ISBN 978-0-231-53868-8 (ebook) 1. Health insurance. 2. Moral hazard. 3. Risk (Insurance) I. Title. HG9383.F56 2014 368.38'2—dc23 2014011303

Columbia University Press books are printed on permanent and durable acid-free paper. This book is printed on paper with recycled content. Printed in the United States of America c 10 9 8 7 6 5 4 3 2 1 Cover design: Noah Arlow References to websites (URLs) were accurate at the time of writing. Neither the author nor Columbia University Press is responsible for URLs that may have expired or changed since the manuscript was prepared.

CONTENTS

Foreword by Joseph E. Stiglitz vii

INTRODUCTION

1

JOSEPH P. NEWHOUSE

MORAL HAZARD IN HEALTH INSURANCE: DEVELOPMENTS SINCE ARROW (1963) 13 AMY FINKELSTEIN

COMMENTARY

45

JONATHAN GRUBER

COMMENTARY

55

KENNETH J. ARROW

COMMENTARY

65

JOSEPH E. STIGLITZ

DISCUSSION

73

ARROW (1963): UNCERTAINTY AND THE WELFARE ECONOMICS OF MEDICAL CARE 81 KENNETH J. ARROW

Notes on Contributors 137 Index 140

FOREWORD JOSEPH E. STIGLITZ

T

his book grew out of the fifth annual Kenneth J. Arrow Lecture that took place in April 2012. This series honors one of Columbia’s most distinguished graduates, Kenneth J. Arrow. Each year’s lecture focuses on the intellectual developments that have arisen from one of his many seminal contributions to the discipline. Over the past six years, the lecture series has covered a huge variety of topics, just as Ken’s work has. The books in the Kenneth J. Arrow Lecture Series are all based on one of these lectures. The first lecture, which Bruce Greenwald and I delivered, followed up on Ken’s 1962 paper on learning by doing.1 That lecture led to a book published in 2014 by Columbia University Press, Creating a Learning Society: A New Approach to Growth, Development, and Social Progress. The second Arrow lecture, by Nobel Prize winners Eric Maskin and Amartya Sen,2 focused on Ken’s Ph.D. thesis, published as Social Choice and Individual Values in 1951. Maskin and Sen’s lecture formed the basis of the second

FOREWORD

book in the series, The Arrow Impossibility Theorem (2014). The third book, Speculation, Trading, and Bubbles (2014), was based on a lecture by José A. Scheinkman. Scheinkman considered Arrow’s fundamental work on general equilibrium, where he introduced the concept of what is now called Arrow-Debreu securities, revolutionizing thinking about financial markets. The fourth lecture was on global warming, a subject about which Arrow has felt passionately. He served with me on Working Group III of the 1995 Assessment of the International Panel on Climate Change (IPCC), and we wrote a paper, together with several of our other colleagues in the working group, on the effect of uncertainty on the appropriate rate of discounting, to take into account the future costs of global warming.3 Sir Partha Dasgupta’s lecture, “Persons and Time in the Welfare Economics of Climate Change,” further pursued the question of the appropriate responses to the threat of global warming. (The sixth Arrow lecture, delivered by Christian Gollier, also considered the economics of discounting, and the books based on his lecture and on Dasgupta’s are forthcoming.) This volume presents the fifth annual Kenneth J. Arrow lecture, by Amy Finkelstein, one of the country’s leading health economists and, like Ken, winner of the John Bates Clark Award (Arrow received this honor in 1957; Finkelstein, in 2012). It picks up on another classic Arrow paper, “Uncertainty and the Welfare Economics of Medical Care,” published in 1963 (and reprinted in this volume), in which he analyzed the implications of moral hazard—a concept that had not yet entered into the mainstream economics Q

viii

Q

FOREWORD

literature—in the context of health insurance. One of the reasons that the standard competitive model—the formulation and analysis of which he himself had played such a pivotal role in—did not provide a good description of the insurance market is moral hazard, the fact that the provision of insurance attenuates incentives to avoid the insured-against event.4 (Another is adverse selection. His 1965 Jahnsson lecture, “Aspects of the Theory of Risk Bearing,” highlighted the importance of both of these.) His 1963 paper was a seminal contribution both to the economics of (health) insurance and to the theory of moral hazard. It launched two huge bodies of literature. We were pleased that one of the major young contributors to that literature accepted our invitation to deliver the fifth Arrow lecture. The questions Ken raised are still front and center in one of today’s central policy issues, the provision of health insurance. Prior to the passage of the Affordable Care Act (commonly called Obamacare), more than 50 million5 Americans lacked health insurance. America’s poor health performance is partially the result of this deficiency. The talk provided an opportunity to connect some of these policy issues with basic economic research. Amy Finkelstein is Professor of Economics at MIT, CoDirector of the Public Economics Program at the National Bureau of Economic Research, and Co-Editor of the Journal of Public Economics, which I helped found many years ago. Her research in public finance and health economics focuses on market failures and government intervention in insurance markets and the impact of public policy in the healthcare sector. Q

ix

Q

FOREWORD

This volume includes not only Amy’s lecture but also the commentaries of its discussants. The first is Jonathan Gruber, who is also a Professor of Economics at MIT and CoDirector of the Healthcare Program at the National Bureau of Economic Research. He has been involved in health economics for a very long time and has won a number of awards for his work, including an Inaugural Medal for Best Health Economist Age Forty and Under from the American Society of Health Economists in 2006, and the National Institute for Health Care Management Foundation Health Care Research Award in 2012. He has also served as Deputy Assistant Secretary for Economic Policy at the Treasury Department. He did important work on the Massachusetts Healthcare Reform, which was the prototype for the Affordable Care Act. I served as the second discussant. Our third discussant is Ken Arrow himself. One of the wonderful aspects of the Arrow Lecture is that Ken graciously agrees to comment on the talk, and that is always an exciting moment, because it reflects both how he thought about the problem many years ago and how thinking about these debates has evolved over subsequent decades. Joseph P. Newhouse has also graciously provided an introduction to this volume, which provides context for the ensuing discussion. Newhouse was one of the leaders of the famous RAND Health Insurance Experiment, and of the landmark Oregon Health Insurance Experiment, which Amy discusses later in this book. In 2014, he received the Victor R. Fuchs Lifetime Achievement Award from the American

Q

x

Q

FOREWORD

Society of Health Economists, in recognition of his pioneering work in the field. I want to acknowledge Columbia University Press, the Program for Economic Research, and the Committee on Global Thought for their support of the Arrow Lecture Series, and Laura Morrison, Sasha de Vogel, and Robin Stephenson for helping organize the event. I also want to especially thank Ken for honoring us again with his attendance, and Amy and Jon for sharing their work and thoughts. 1. Arrow, Kenneth J. “The economic implications of learning by doing.” The review of economic studies (1962): 155–173. 2. Ken Arrow received the Nobel Prize in 1972; Amartya Sen, in 1998, and Erik Maskin, in 2007. 3. Published as Arrow, K. J., W. R. Cline, K. G. Mäler, M. Munasinghe, R. Squitieri, and J. E. Stiglitz, 1996: “Discounting.” In: Climatic Change 1995: Economic and Social Dimensions of Climate Change, Second Assessment of the Intergovernmental Panel on Climate Change [Bruce, J. P., H. Lee, and E. F. Haites (eds.)]. Cambridge University Press, Cambridge and New York, pp. 129–144. 4. An extreme form of moral hazard is that where the individual actually takes action to precipitate the insured-against event. 5. See U.S. Census Bureau. “Income, Poverty, and͒Health Insurance Coverage in the United States: 2009,” Table 8, Issued in September 2010.

Q

xi

Q

MORAL HAZARD IN HEALTH INSURANCE

INTRODUCTION JOSEPH P. NEWHOUSE

E

ach academic year, I teach the first session of a onesemester course in health economics for second-year graduate students. The reading for that session consists of two seminal works in health economics: Kenneth Arrow’s “Uncertainty and the Welfare Economics of Medical Care” and Michael Grossman’s The Demand for Health: A Theoretical and Empirical Investigation (Arrow 1963; Grossman 1972).1 These two works have resulted in two largely nonoverlapping streams of the by-now vast health economics literature. Arrow’s article led, somewhat belatedly, to a literature on the functioning of markets for medical services and health insurance.2 Grossman’s book led to a literature on determinants of the health status of the population, only one determinant of which, as both Arrow and Grossman emphasized, is medical care. The real-world context for Arrow’s 1963 article was the growing clinical capabilities of medical care and the concomitant acceleration of medical care’s claims on the economy’s resources.3 The economics context was general equilibrium

INTRODUCTION

theory, which considered the existence and properties of competitive equilibrium (Arrow and Debreu 1954). Among the stringent requirements for the efficiency of competitive markets were the existence of a market for all products, including markets for risk, and the availability of information about costs and benefits. Arrow focused on the shortcomings of medical markets, especially incomplete markets for risk bearing and the imperfect information available to doctors, patients, and potential insurers. Arrow emphasized that markets to insure all medical risks could not exist, in part because what a consumer sought from visiting a physician was information about the cause of a medical problem and its appropriate treatment. Because it was practically impossible to enumerate all possible contingencies and a fixed amount the policy would pay for each, an insurance contract specifying a payoff could not exist. In other words, markets to insure against these risks did not exist, and so an efficient outcome was unattainable. Furthermore, Arrow argued that nonmarket institutions such as reliance on professional ethics had arisen to mitigate the welfare effects of the lack of markets against all risks. Because these nonmarket institutions were found in many societies, one could presume that they were welfare enhancing. Arrow’s attention to information and uncertainty gave rise to two different literatures within the economics of medical care. The first literature took up the effects of selection in individual or small-group insurance markets; two of the first articles in this literature are the famous papers of George Akerlof and of Michael Rothschild and Joseph Stiglitz, later recognized by Nobel Prizes for Akerlof and Stiglitz (Akerlof Q

2

Q

INTRODUCTION

1970; Rothschild and Stiglitz 1976). The second literature took up moral hazard, which is the focus of Amy Finkelstein in the next chapter of this book. The part of Arrow’s paper discussing moral hazard was interpreted by Mark Pauly in a comment on Arrow’s article as the response of the demand for medical care to its out-of-pocket price (Pauly 1968). Pauly’s comment is surely the most famous and most often cited comment in the entire health economics literature. I personally became interested in empirically assessing the degree of moral hazard the year I finished graduate school, which happened to be the same year Pauly’s comment and Arrow’s response was published (Arrow 1968). Out of this interest grew the RAND Health Insurance Experiment, which I led and to which Amy refers in the next chapter. Two additional papers by Ken Arrow that generalized a result proved in the appendix of his original 1963 paper (Arrow 1974, 1976)4 were a little-known part of the RAND experiment work. The original result showed that if insurance was not actuarily fair, as it never was in reality, the optimal health insurance contract was full insurance above a deductible. This result, however, assumed both that utility was not state dependent and that the health shock being insured against resulted in a fixed monetary loss. One of Arrow’s generalizations allowed for a consumer’s utility to depend on his or her health state and yielded an unsurprising result: Depending on how marginal utility varied with any change in health status, the optimal deductible would vary. Notably, Amy herself has made an empirical contribution to this question, suggesting that state dependence is on average negative, meaning that health shocks tend Q

3

Q

INTRODUCTION

to reduce the marginal utility of money at a given level of wealth—another reason, in addition to administrative costs, that insurance should not be complete (Finkelstein, Luttmer, and Notowidigo 2009). The second generalization stemmed from the observation that actual insurance policies in the 1960s generally did not resemble Arrow’s optimal policy; for example, they frequently had co-insurance in addition to a deductible.5 Arrow therefore took up the welfare consequences of this type of insurance contract. He pointed out that as the co-insurance rate increased, the individual had a greater incentive to economize on medical care but also bore more risk, a tradeoff later emphasized by Richard Zeckhauser (1970). At that time, however, rather little was known about the strength of the incentive to economize as the co-insurance rate changed, or the degree of moral hazard. That question remained to be investigated both as an empirical issue in health economics and as a normative issue concerning the optimal amount of cost-sharing in insurance policies. Amy gives an excellent discussion of these issues in the next chapter. In the remainder of this introduction, I wish to make three points about the moral hazard literature that Amy discusses. Textbook discussions of this subject typically show a demand curve, usually linear; introduce insurance into the diagram as a subsidy; define and illustrate deadweight loss; and give a summary number such as –0.2 for the price elasticity. In my view, there are three ways to improve on this conventional discussion. The first simply concerns exposition; the other two are substantive issues about the welfare consequences of moral hazard. Q

4

Q

INTRODUCTION

Economists prefer elasticity as a measure of how demand responds to a change in price because it is independent of the units in which price and quantity are measured, thereby allowing elasticities to be compared across markets. And, as Amy notes, one of the conclusions of the empirical literature on the demand for medical care is that the price elasticity of demand is around –0.2. While elasticity is certainly the accepted way to present a measure of responsiveness to price in the economics literature, I do not think it is helpful in the health insurance context and can in fact mislead. Health insurance often reduces out-of-pocket prices at the point of service to rather small amounts. Indeed, there is still an active debate, especially outside the United States, about whether medical services should be free at the point of service, and in a number of countries they are. In the often relevant context of zero or low-cost sharing, I think it is preferable to illustrate the response to changes in cost-sharing as simply the percentage change in quantity from a given change in cost-sharing rather than as an elasticity. The arc elasticity implied by a change from zero co-insurance (free care) to some positive level of co-insurance is by definition always half the percentage change in quantity. Thus, in this extreme but relevant case, the change in quantity is a sufficient statistic for the elasticity; the elasticity adds no further information. Even in less extreme cases, however, such as a change in a drug co-payment from $5 to $10, the percentage change in price is sufficiently large that even a nontrivial induced percentage change in quantity results in a seemingly small elasticity. Q

5

Q

INTRODUCTION

For example, to many economists an elasticity of –0.2 implies little price responsiveness. But if a change from zero co-insurance to some positive rate of co-insurance is stated as a 40 percent reduction in demand rather than as an elasticity of –0.2, most individuals would consider that a large response. Estimates of elasticity usually come from noninsured contexts, and so economists’ intuition about elasticity may mislead them in the health insurance context. For that reason, when presenting the results of the RAND experiment, I personally present the percentage response in demand at a given co-insurance rate relative to demand when care is free rather than the implied arc elasticity. Not only do nontechnical audiences find this a more understandable number, but even economists come away with the conclusion that choices about the amount of cost-sharing really do matter.6 This is the first improvement on the conventional textbook discussion that I can suggest. A second, much more substantive issue concerns the normative meaning of the demand curve. Discussions of deadweight loss in health economics textbooks generally assign the demand curve its usual normative meaning as indicating the rate at which the marginal consumer would trade off medical care against other goods and services; indeed, that is implied in the definition of deadweight loss and is central to Pauly’s comment on Arrow’s seminal paper (Pauly 1968). Others, citing agency issues of physicians, assign no normative meaning to the demand curve; Robert Evans was an early exponent of that view (Evans 1974). Arrow, in his response to Pauly, took a more nuanced view, though he was clear that he thought there was an agency issue and that Q

6

Q

INTRODUCTION

one should not interpret the demand curve in the usual way (Arrow 1968). At first blush, the results from the RAND experiment seem to support the usual textbook view. More complete insurance substantially increased the demand for care, with little or no apparent effect on the average person’s health, although low-income individuals with elevated blood pressure had their blood pressure less well controlled—with nontrivial implications for their predicted downstream mortality (Newhouse and the Insurance Experiment Group 1993). This appears to demonstrate that the additional services received by most of those whose care was free had little value to them, consistent with the standard discussion of deadweight loss. Such an interpretation of the RAND results, however, is too simple. Another finding of the RAND experiment was that variation in out-of-pocket price affected the use of both efficacious and nonefficacious medical services. But if cost-sharing led to a reduction in the use of efficacious services, why was there no effect on health for the average person? Although measurement error can always potentially account for null findings, confidence intervals for the lack of health effects were tight. Rather than measurement error, I believe the main explanation is that the additional services that those with free care received had positive effects on health for some and negative effects for others; this is consistent with finding positive effects among low-income, chronically ill patients where the odds of additional medical care having a positive effect are higher. It is also consistent with all participants in the RAND experiment being under 65 years of Q

7

Q

INTRODUCTION

age and living in the community, meaning that most of them were in good health. As a result, the odds that a service at the margin helped them were probably offset by the odds that it hurt them. I have felt more confidence in this explanation over time as evidence of medical error and poor quality of care has piled up (Institute of Medicine 1999, 2001; McGlynn et al. 2003). If the explanation is true, however, it causes a problem for the usual textbook treatment of deadweight loss from moral hazard, as the usual interpretation implies the consumer should never consume services with an expected negative value: textbook demand curves do not extend below the horizontal axis. A third way to improve on the standard textbook discussion is to recognize cross-price effects, especially whether lower prices for some services may reduce overall spending; for example, if lower co-payments for drugs lead to greater compliance with a drug regimen that in turn reduces downstream hospitalizations (Newhouse 2006; Ellis and Manning 2007; Goldman and Philipson 2007; Choudhry et al. 2011; Glazer and McGuire 2012). Such an effect, however, is somewhat inconsistent with conventional welfare economics. Lack of compliance can in some cases be rationalized by pointing to costs of complying with the prescribed medical regimen, such as a drug with severe side effects, but in many cases behavioral economic explanations such as hyperbolic discounting are a more plausible explanation of the lack of compliance. The latter explanation, however, fits awkwardly with the textbook discussion of deadweight loss. In the next chapter, Amy gives an admirable summary of the current state of the literature on moral hazard. Fortunately Q

8

Q

INTRODUCTION

for those now embarking on a career in health economics, many questions relating to moral hazard remain unanswered. One of the more important is whether selective reductions in the out-of-pocket or after-insurance price of certain medical services would be welfare enhancing by inducing greater use of medically efficacious services and reduced use of inefficacious services (Pauly and Blavin 2008). Obtaining answers will be tedious and will probably not be considered a large intellectual breakthrough, but nonetheless will potentially be of great social value. Another considers the agency questions about how physicians and other medical care providers respond to the amount and basis of remuneration for their services such as fee-for-service or capitation. The degree to which physicians and physician groups respond to financial incentives bears on how they should be paid, including the optimal amount of risk for a physician group and what relative prices should be in an administered relative price system such as the Medicare Fee Schedule. This can be viewed as a moral hazard problem, albeit on the supply side (Ellis and McGuire 1986). As countries everywhere struggle with obtaining greater value from the money they spend on health care, they will be paying attention to credible results on both these questions.

NOTES 1. Arrow’s paper is his second most frequently cited paper in the ISI Citation Index, trailing only his 1962 paper on learning-bydoing (Arrow 1962). 2. As an aside, “health” insurance appears inaptly named as it actually provides financial protection against the cost of medical Q

9

Q

INTRODUCTION

3. 4.

5.

6.

services and does not insure health, but “medical care insurance” does not roll off the tongue as easily as “health insurance.” The enactment of Medicare and Medicaid in 1965 only increased the article’s relevance. Both these publications are also available through the RAND website (www.rand.org) as RAND Report numbers R-1108 and R-1281. An even more striking departure from optimal insurance was upper limits on the insurer’s payout, either an annual or a lifetime limit or both. The Affordable Care Act of 2010 made such limits illegal in the United States. There is the further potential problem in giving a summary value for price elasticity that the demand function may not exhibit a constant elasticity; in particular, a linear demand curve does not have constant elasticity.

REFERENCES Akerlof, George. 1970. “The Market for ‘Lemons’: Qualitative Uncertainty and the Market Mechanism.” Quarterly Journal of Economics 74(3):488–500. Arrow, Kenneth J. 1962. “The Economic Implications of Learning-byDoing.” Review of Economic Studies 29(3):155–73. Arrow, Kenneth J. 1963. “Uncertainty and the Welfare Economics of Medical Care.” American Economic Review 53(5):941–73. Arrow, Kenneth J. 1968. “The Economics of Moral Hazard: Further Comment.” American Economic Review 58(3, part 1):537–9. Arrow, Kenneth J. 1974. “Optimal Insurance and Generalized Deductibles.” Scandanavian Actuarial Journal 1:1–42. Arrow, Kenneth J. 1976. “Welfare Analysis of Changes in Health Coinsurance Rates.” In The Role of Health Insurance in the Health Services Sector, ed. Richard N. Rosett. New York: Neale Watson Aademic Publications.

Q

10

Q

INTRODUCTION

Arrow, Kenneth J., and Gerard Debreu. 1954. “Existence of an Equilibrium for a Competitive Economy.” Econometrica 22(3):245–90. Choudhry, Niteesh K., Jerry Avorn, Robert J. Glynn, et al. 2011. “Full Coverage for Preventive Medications after Myocardial Infarction.” New England Journal of Medicine 365(22):2088–97. Ellis, Randall P., and Thomas G. McGuire. 1986. “Provider Behavior under Prospective Reimbursement.” Journal of Health Economics 5(2):129–51. Ellis, Randall P., and Willard G. Manning. 2007. “Optimal Health Insurance for Prevention and Treatment.” Journal of Health Economics 26(6):1128–50. Evans, Robert G. 1974. “Supplier Induced Demand: Some Empirical Evidence and Implications.” In The Economics of Health and Medical Care (pp. 162–73), ed. Mark Perlman. New York: Macmillan. Finkelstein, Amy N., Erzo Luttmer, and Matthew Notowidigo. 2009. “Approaches to Estimating the Health State Dependence of the Utility Function.” American Economic Review 99(2):116–21. Glazer, Jacob, and Thomas G. McGuire. 2012. “A Welfare Measure of ‘Offset Effects’ in Health Insurance.” Journal of Public Economics 96(5–6):520–23. Goldman, Dana P., and Tomas J. Philipson. 2007. “Integrated Insurance Design in the Presence of Multiple Medical Technologies.” American Economic Review 97(2):427–32. Grossman, Michael. 1972. The Demand for Health: A Theoretical and Empirical Investigation. New York: Columbia University Press. Institute of Medicine. 1999. To Err Is Human. Washington, D.C.: National Academy Press. Institute of Medicine. 2001. Crossing the Quality Chasm: A New Health System for the 21st Century. Washington, D.C.: National Academy Press. McGlynn, Elizabeth, Steven M. Asch, John Adams, Joan Keesey, Jennifer Hicks, Alison DeCristofaro, Eve A. Kerr. 2003. “The Quality of Health Care Delivered to Adults in the United States.” New England Journal of Medicine 348(26):2635–45.

Q

11

Q

INTRODUCTION

Newhouse, Joseph P. 2006. “Reconsidering the Moral Hazard-Risk Aversion Tradeoff.” Journal of Health Economics 25(5):1005–14. Newhouse, Joseph P., and the Insurance Experiment Group. 1993. Free For All: Lessons from the Health Insurance Experiment. Cambridge, Mass.: Harvard University Press, 1993. Pauly, Mark V. 1968. “Comment.” American Economic Review 58(3): 531–7. Pauly, Mark V., and Frederic E. Blavin. 2008. “Moral Hazard in Insurance, Value-Based Cost Sharing, and the Benefits of Blissful Ignorance.” Journal of Health Economics 27(6):1407–17. Rothschild, Michael, and Joseph Stiglitz. 1976. “Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information.” Quarterly Journal of Economics 90(4):629–50. Zeckhauser, Richard J. 1970. “Medical Insurance: A Case Study of the Tradeoff between Risk Spreading and Appropriate Incentives.” Journal of Economic Theory 2(1):10–26.

Q

12

Q

MORAL HAZARD IN HEALTH INSURANCE Developments Since Arrow (1963) AMY FINKELSTEIN

I

was honored to give the fifth annual Kenneth J. Arrow Lecture at Columbia University in April 2012. When I was first asked to give the lecture in honor of Ken Arrow and on the topic of the economics of health care, I thought that this would be simple. Ken wrote a seminal 1963 paper titled “Uncertainty and the Welfare Economics of Medical Care.” My plan was to talk about how this paper has shaped and influenced the subsequent field of health economics, a field that did not exist when he wrote the paper. It is striking to re-read the paper: One quickly realizes that virtually everything that has happened in health economics in the past half century can trace its origins back to that paper. But that’s also when I realized that I had a real problem: I only had 50 minutes to talk about the wide-ranging impact of the paper . . . and even I can’t talk quite that fast! So what I decided to do instead was to focus on just one aspect of Arrow’s original paper, which relates to a topic I have been researching for the past several years: the economics of moral hazard.

MORAL HAZARD IN HEALTH INSURANCE

To put this topic in perspective, there are essentially two central facts of the U.S. health care system. The first is that approximately 46 million people, or about 15 percent of the population, lack health insurance. The second is that health care spending is a large and growing share of our economy. In 1960, only 5 percent of GDP was spent on health care; in 2012, that share was 17 percent. That has real implications for the public sector, because in 2012 spending on health care was 20 percent of the federal budget and a substantial share of state budgets. Moreover, the major driver of projected federal spending growth is from public health insurance, specifically Medicare and Medicaid (Congressional Budget Office 2012). Not surprisingly, the recent health care reform—the Affordable Care Act of 2010—tried to adjust both these issues. It focused on trying to cover the uninsured and to rein in the growth of health care spending. But there is a fundamental tension in these goals, in that by itself accomplishing the first goal exacerbates the second problem. All else equal, insuring the uninsured is likely to increase health care spending. Why is that? It has to do with the moral hazard effects of health insurance, which is the subject of this essay. To my knowledge, Arrow’s 1963 paper is the first academic reference to moral hazard and health insurance. Arrow defined moral hazard in health insurance as the concept that “medical insurance increases the demand for medical care” (Arrow 1963:961). At the end of the introduction in that article, he lays down a gauntlet for subsequent researchers. He says (and I really like this quotation):

Q

14

Q

MORAL HAZARD IN HEALTH INSURANCE

The discussion is not designed to be definitive, but provocative. In particular, I have been chary about drawing policy inferences; to a considerable extent, they depend on future research, for which the present paper is intended to provide a framework. (948)

And indeed it did so. What I want to explore here is how we have risen to this challenge that he laid down in this article 50 years ago. In particular, I will describe two separate empirical challenges posed by his discussion of moral hazard. The first is an existence question: Is the idea of moral hazard, which is an interesting theory, empirically relevant? Is the demand for medical care really price-sensitive? And the second relates to the challenge he poses about drawing policy inferences: How to estimate the likely impact of alternative health insurance policies or contracts on both the level and the growth of health care spending?

IS DEMAND FOR MEDICAL CARE REALLY PRICE-SENSITIVE? What do we mean when we say moral hazard and health insurance? Arrow defined it as medical insurance increasing the demand for medical care. Now why would that happen? There are essentially two ideas in the literature. The first, due to Ehrlich and Becker (1972), has come to be known as ex ante moral hazard. This is the idea that if I have health insurance and it will pay my medical bills when I get sick,

Q

15

Q

MORAL HAZARD IN HEALTH INSURANCE

then I have less incentive to invest in maintaining my health, because when I get sick, the financial consequences will now be borne by someone else; therefore, I am going to eat, drink, and be merry. I might smoke more, drink more, exercise less, et cetera. The second idea, which was formalized by Mark Pauly (1968), has come to be known as ex post moral hazard. For this idea, let’s forget that health insurance may affect my investments in my health; let’s just take my health as given. Ex post moral hazard is the idea that, at a given level of health, I am going to choose to consume more medical care because the price of that medical care is lower. This is basically about a demand curve and the price sensitivity of demand for medical care. For the most part, the literature has tended to adopt the second definition. Moral hazard and health insurance have come to mean price sensitivity of demand for medical care, rather than the impact of health insurance on investment in one’s health. For the remainder of this text, I’ll focus on the price sensitivity of demand. The first question is whether demand for medical care is in fact sensitive to the price of care. On the one hand, this certainly seems like a natural property of any demand curve: If you lower the price of something, people buy more of it. There is an alternative view, however, which holds that medical care is determined not by price but by needs. One summary of this is what I call the rhetorical case against the notion of moral hazard and health insurance. It was put forward in a relatively recent New Yorker article by Malcolm Gladwell titled “The Moral-Hazard Myth”: Q

16

Q

MORAL HAZARD IN HEALTH INSURANCE

The moral-hazard argument makes sense . . . only if we consume health care in the same way that we consume other consumer goods, and to economists like [ John] Nyman this assumption is plainly absurd. We go to the doctor grudgingly, only because we’re sick. “Moral hazard is overblown,” the Princeton economist Uwe Reinhardt says. “You always hear that the demand for health care is unlimited. This is just not true. . . . Do people really like to go to the doctor? Do they check into the hospital instead of playing golf ?” (Gladwell 2005)

If that is the rhetorical case, what exactly is the evidence? Ultimately, it is an empirical question: When people are given health insurance, do they consume more medical care and thus increase health care spending? Empirically, it is somewhat challenging to get at this. We do not just want to compare people with and without insurance and look at whether people with insurance spend more than people without insurance, because people with insurance are different than those without insurance. They have different income, different employment status, and different health. They are different in ways that are likely correlated with the demand for health care; basic adverse selection theory suggests that individuals who are less healthy will be more likely to purchase health insurance (Akerlof 1970; Rothschild and Stiglitz 1976). In fact, if we just do these cross-tabulations in the data, we will quickly discover what looks like moral hazard: People with health insurance spend more on health care than people without it. We’ll also quickly discover that it looks Q

17

Q

MORAL HAZARD IN HEALTH INSURANCE

like health insurance kills people, because people with health insurance have higher mortality than people without health insurance. So that might give us some pause in what we can infer about the impact of health insurance from these types of observational comparisons. The ideal solution to test the null hypothesis that Malcolm Gladwell and others have put forward—that there is no moral hazard or no price sensitivity of the demand for medical care—would be a randomized controlled trial in which different insurance is randomly assigned across individuals. Then the selection problem does not exist, and this allows a researcher to infer what the effect is of giving insurance to one group and not another.

THE OREGON MEDICAID EXPERIMENT Remarkably, in the United States there have been two randomized controlled trials on health insurance. (You can think it’s remarkable either that there have been “only two” randomized evaluations of such an important topic or that we have had any randomized trials at all.) One of the experiments was the RAND Health Insurance Experiment from the 1970s. The other is the Oregon Health Insurance Experiment, which I have been leading with Kate Baicker at the Harvard School of Public Health. It is based on a Medicaid lottery conducted in Oregon in 2008, and our analysis of it is ongoing. We of course have more details available for those who are interested (see, e.g., Finkelstein et al. [2012] and also our study website, www.nber.org/oregon). Q

18

Q

MORAL HAZARD IN HEALTH INSURANCE

Let me describe the Oregon experiment, and then one can judge the evidence for oneself. Medicaid is public health insurance for the indigent. In Oregon, the state has an expansion program to cover people who are financially but not categorically eligible for Medicaid. Essentially, these are lowincome people, below 100 percent of the federal poverty line. This is extreme poverty—less than $10,000 in annual income for a single person. These adults are uninsured but they are able-bodied, so they can’t enroll in Medicaid because they are on welfare or they’re disabled, and they are not otherwise eligible for Medicaid. Oregon had money to cover some of the low-income, uninsured adults in this group, but not all of those that are eligible. The state only had money to cover ten thousand people, so it was necessary to think about how best to do this. State officials decided it was not fair to award the limited slots on a first-come, first-served basis because that would privilege people who are better connected. Federal law actually prohibits awarding Medicaid on the basis of health status, so the state decided the only equitable thing to do was to run a lottery. The state sponsored a big public relations campaign and asked interested individuals to sign up for the lottery. About ninety thousand lowincome adults signed up, and the state randomly drew about thirty thousand names to be eligible to apply for Medicaid. We have results now from the first year of this ongoing experiment. This is just one of many pieces of data and is shown to give a sense of the matter. Figure 1 looks at the probability of hospital admission over about a 16-month period after the lottery. The black in the figure shows the Q

19

Q

MORAL HAZARD IN HEALTH INSURANCE

10

Percentage

8

6

4

2

0

All

Via emergency department

Not via emergency department

Control mean Control mean plus Medicaid effect CI for Medicaid effect

Figure 1 Probability of Hospital Admission. These hospital discharge data show the impact of Medicaid on hospital admissions, using evidence from the Oregon Health Insurance Experiment. Outcomes were measured over an approximately 1-year period. Source: Finkelstein et al. (2012).

probability of hospital admission for the controls, those who did not win the lottery. (We try to avoid referring to them as “losers.”) They have about a 7 percent admission rate over a 16-month period. And the gray in the figure shows the implied increase in the probability of going to the hospital due to Medicaid. One can see that it’s about two percentage points, or 30 percent, higher. Q

20

Q

MORAL HAZARD IN HEALTH INSURANCE

This empirical evidence from a randomized controlled trial contradicts the notion that demand for medical care is not price sensitive—that because no one wants to go to the hospital, they’re not going to do it even if it’s less expensive. There is an interesting effect if one looks at the other two bars in figure 1, where we broke up hospital admissions into those that come through the emergency department and those that do not. Perhaps not surprisingly, what one sees is that almost all of the increase in hospital admissions consists of admissions that do not come through the emergency department; these are presumably ones that are less serious or less time sensitive and, therefore, perhaps not surprisingly, more price sensitive. It is a fairly intuitive result. We have much more evidence from just this first year. We see that Medicaid increases not only hospital admissions, as was just demonstrated, but also the probability of taking prescription drugs and of going to the doctor. Upon tallying the cost of all of these things, we estimate that Medicaid increases annual medical spending by about 25 percent. People on Medicaid, when they are randomly assigned to it, are spending 25 percent more—which is about $750 a year for this population—than those who are uninsured. This is fairly compelling and definitive evidence rejecting the null hypothesis of no spending effective health insurance, or no moral hazard. As an aside, we do not find, at least in the first year of the Oregon experiment, any evidence of ex ante moral hazard, where someone might say, “Now that I have health insurance, I don’t have to take as good care of myself.” For example, we did not find any change in smoking behavior. Now, Q

21

Q

MORAL HAZARD IN HEALTH INSURANCE

this may be because the theoretical possibility that when one has health insurance, one is less concerned about one’s health is not actually operative, or it might be operative and counterbalanced by the fact that people with health insurance go to the doctor more, and the doctor urges them not to smoke. That is something that remains to be sorted out, but it is consistent with the intuition that there is not much of this ex ante effect.

THE RAND HEALTH INSURANCE EXPERIMENT I also want to mention briefly the other health insurance experiment, the famous RAND Health Insurance Experiment from the 1970s. Here about six thousand individuals were randomly assigned to plans with different levels of consumer cost-sharing. Unlike the Oregon experiment, where we compared the impact of public insurance relative to having no insurance, everyone in the experiment receives private insurance, but the private insurance in different experimental arms differs in terms of how much consumer cost-sharing it requires. The RAND experiment was led by Joseph Newhouse, one of our collaborators in the Oregon experiment, and it is summarized very nicely in his 1993 book Free for All (Newhouse and the Insurance Experiment Group 1993). I wrote a shorter summary piece on the RAND experiment with some collaborators that is designed to be accessible to advanced undergraduates (Aron-Dine, Einav, and Finkelstein 2013).

Q

22

Q

MORAL HAZARD IN HEALTH INSURANCE

In the RAND experiment, families were randomly assigned to private insurance plans with different amounts of consumer cost-sharing. The most comprehensive coverage was called the free-care plan: This had zero cost-sharing, so zero was what the consumer had to pay out-of-pocket. Then there were various positive consumer cost-sharing plans where the consumer had to pay something out-ofpocket, depending on what one was assigned: maybe one had to pay 25 percent of the cost out-of-pocket, or 50 percent, or 95 percent. All the plans had low out-of-pocket maximums, or stop losses, so that even if a participant had to pay 95 percent out-of-pocket, he or she had to pay that only up to some maximum out-of-pocket amount, which was at most $1,000 (in 1970 dollars), after which the insurance covered everything. Even the people in the least generous plan still had catastrophic coverage. In fact, about a third of families were hitting the stop loss, so it was quite binding. Once again, the evidence of moral hazard was compelling—spending is lower when consumer cost-sharing is higher, and when consumers have to pay a higher share of the cost out-of-pocket, they spend less. To give just one number: if we look at people assigned to the plan with 95 percent consumer cost-sharing—so they have to pay 95 percent of their medical costs up to the stop loss—we find that their annual medical spending is almost two-fifths less than the annual spending for those assigned to the free-care plan. Again this is compelling evidence against the null hypothesis of no moral hazard effects.

Q

23

Q

MORAL HAZARD IN HEALTH INSURANCE

HOW WILL ALTERNATIVE HEALTH INSURANCE POLICIES AFFECT HEALTH CARE SPENDING? So what does this mean? Thus far, what I have hopefully convinced you of is that contrary to what uninformed readers of the New Yorker might believe, moral hazard and health insurance is not a myth. There is compelling empirical evidence from randomized controlled trials that when people are given health insurance, they consume more medical care, and when they are given more comprehensive insurance, they consume even more medical care. But key challenges remain for policy. This goes back to what Arrow talked about in his original article: What are the policy implications of moral hazard in health insurance? In other words, one question that comes up is how do we translate these experimental treatment effects—such as the differences in spending across plans that we see in the RAND experiment—into economic objects of interest? In particular, we are always very interested in how one can take these estimates from the RAND Health Insurance Experiment of what people spent in the specific plans they were randomly assigned to and use those to forecast out-of-sample the likely effects of different plans or different policies that we might be considering today. So I want to devote the second half of this essay to describing how to translate the economics of moral hazard into policy. I want to discuss several key conceptual challenges that immediately confront us if we try to predict the spending effects of alternative health insurance policies or health

Q

24

Q

MORAL HAZARD IN HEALTH INSURANCE

insurance regimens, some of which I have been working on for the past few years. It’s often helpful just to have a specific example to ground discussion and focus ideas, so I am going to describe how to think about the impact of high-deductible health insurance plans on spending. High-deductible insurance plans were encouraged by the Health Savings Accounts Act of 2003, which encouraged people through tax subsidies to buy veryhigh-deductible plans. Within the deductible, a participant pays 100 percent out-of-pocket. Qualifying plans often have deductibles of, say, $3,000, but then the participant receives catastrophic coverage for expenditures beyond that deductible. One is insured against very large expenditures but first has to pay the deductible amount out-of-pocket. Again going back to Arrow’s paper, perhaps not coincidentally this design is actually the theoretically optimal design of a contract when there are risk-averse individuals and concerns about moral hazard. In fact, a goal of this legislation was to get people into high-deductible plans to reduce the level and growth of health spending, while still providing catastrophic coverage that is so valuable to risk-averse individuals. There are at least three questions that come up when we want to think about how a high-deductible health insurance plan is going to affect health spending: 1. Which price matters to consumers? Is it the deductible that they face at the start of the year or their anticipation that by the end of the year they may have spent past the

Q

25

Q

MORAL HAZARD IN HEALTH INSURANCE

deductible? Which are they going to respond to in their spending behavior? 2. Who is going to select the high-deductible plans, and how is that going to affect the impact of these plans on spending? 3. Why would health insurance affect spending growth, which is really the key problem, as opposed to just the level of health spending?

WHAT PRICE MATTERS TO CONSUMERS? Typical plans—including those in the RAND Health Insurance Experiment and those in the real world, both then and now—are highly nonlinear. Instead of facing a linear relationship between total medical spending and out-of-pocket costs, an individual faces something like what’s shown by the dashed line in figure 2. Initially, the individual is in the deductible range, and so it’s about a 45-degree line there because the individual is paying 100 percent out-of-pocket, so out-of-pocket spending is rising one for one with total spending. At some point, the individual spends past the deductible and into the co-insurance arm. Let’s say the individual faces a 20 percent marginal price in the co-insurance arm, so his or her out-of-pocket spending is rising one-fifth for every dollar spent. Then at some point the individual hits the stop loss—that’s the horizontal line in figure 2—so his or her outof-pocket spending does not move with total spending, and medical care is now free. At the beginning of the year when these plans reset and individuals are faced with a deductible, do they think that Q

26

Q

Out-of-pocket spending

MORAL HAZARD IN HEALTH INSURANCE

Total spending Figure 2 Nonlinear health insurance contracts. What price matters to consumers?

the marginal price of going to the doctor is the full price— the 100 percent that they have to pay out-of-pocket—or do they think about the fact that by the end of the year, maybe they’ll have had a bunch of illnesses anyway and be past that deductible? Which price do they respond to? That’s obviously going to have implications for how much of a spending reduction will be obtained for introducing a high-deductible plan. If people react as if the price is the spot or sticker price of 100 percent, they will reduce spending a lot more than if they forecast that by the end of the year, they may face a much lower effective marginal price. In particular, if individuals are myopic, they might respond to the introduction of a deductible as if their price has increased dramatically to 100 percent. In contrast, if these are fully rational, forward-looking and not liquidity-constrained Q

27

Q

MORAL HAZARD IN HEALTH INSURANCE

individuals, they may recognize that the spot price should not affect their consumption decision, only the end-of-year price. Imagine someone with a $3,000 deductible, but who has a chronic condition that regularly costs more than $10,000 a year. At the beginning of the year, if this person has a headache and is considering whether to go to the doctor to get this unusual headache looked at, this person should not treat the current doctor visit as if he or she bears the full cost out-of-pocket. Rather, he or she should think: “Over the course of the year, I’m naturally going to spend past the deductible anyway and face a marginal price much lower, perhaps zero, of going to the doctor. So the effective marginal price of the current visit is quite low.” The question is, does this person do that? This is something I’ve tried to get at in a recent paper with Aviva Aron-Dine, a graduate student at MIT, and my collaborators Liran Einav and Mark Cullen at Stanford (Aron-Dine et al. 2012), where we asked the question: How important is forward-looking behavior in moral hazard? Like the question of whether moral hazard exists in health insurance, the question of how forward looking people are in their medical consumption is ultimately an empirical question. In this paper, we take advantage of the fact that with employer-provided health insurance, the annual deductible resets on January 1, regardless of when the employee is hired. Whether one is hired in March or in October, the deductible is going to reset in January. As a result, if we imagine someone who is hired into a deductible plan in March and someone else who is hired in October, they both face the same initial spot price of Q

28

Q

MORAL HAZARD IN HEALTH INSURANCE

medical care of one. They have to pay fully for the first doctor visit, but they face very different expected end-of-year prices, because the individual hired in February or March has the whole year to have health shocks accumulate and push him past the deductible, while the individual hired in October or November—or December 31 in the extreme—is very unlikely to get past the deductible. Our basic idea therefore is to use the variation across employees in hiring date to look at a group of people who face the same initial or spot price for medical care but different future prices; those hired later in the year face higher end-of-year expected prices (future prices) because they have less time to spend past the deductible. We therefore compare the initial medical utilization of people hired into the same deductible plan at different months in the year. Of course, we have to take into account that people hired in February may be different than people hired in October and the onset of illness can be different in February than in October. We therefore look at patterns of initial medical use for those hired into a no-deductible plan, which does not have this feature, as a control for seasonal variation in hires or medical spending. What we find is a rejection of the null of complete myopia, or the hypothesis that individuals respond only to the spot price. We find that individuals in a deductible plan hired earlier in the year, who face the same spot price but a lower expected end-of-year price than individuals in the deductible plan hired later in the year, initially use more medical care. In other words, they are more likely to go to the doctor in the first month or two than individuals hired later in the year. Q

29

Q

MORAL HAZARD IN HEALTH INSURANCE

This suggests that individuals are not completely myopic. They do not just respond to the sticker price, so there is some degree of forward-looking behavior, but how much and how important is it? To answer this, we write down and calibrate a simple model. The results suggest that individuals are not fully forward-looking, so they respond both to the spot price and to the future price. What are the implications? With the amount of forward-looking behavior we estimate, we find that the spending reduction of moving from a nodeductible to a high-deductible plan is a lot lower, about 25 to 50 percent, as we would predict if we believed that individuals were fully myopic. The sign of that result is intuitive, and we don’t need to estimate the model to know it. As I described, when individuals are fully myopic, the deductible has the most bite, because they keep thinking that as long as they are in the deductible range, that’s their effective price. So if individuals are not fully myopic, we know the spending reduction from introducing a high-deductible plan will be lower. But the magnitude is also of interest, and we would not have known it without this estimation. I want to provide some context about why it is challenging going from the experiment to these policy implications. The RAND investigators in their original work were aware of all these issues and tried to translate the differences in spending across plans in their experiment into a prediction for how spending would respond in other plans. The result was their now-famous estimate of the price elasticity of demand of –0.2, which is widely used by the Congressional Budget Office and by many others. The RAND investigators Q

30

Q

MORAL HAZARD IN HEALTH INSURANCE

computed this elasticity under the assumption that individuals were fully myopic. And if in fact that’s not the case, the spending effects of some of these plans may be a lot lower than we thought.

WHO SELECTS HIGH-DEDUCTIBLE PLANS? A second and related issue in trying to forecast the likely spending reduction from introducing a high-deductible plan is the question: Who is going to select it? In general, we tend to think of selection and moral hazard as distinct issues. The adverse selection literature has focused on the fact that individuals may differ in their (privately known) underlying health, and that this is going to affect their demand for health insurance. The moral hazard literature has tended to ignore heterogeneity across individuals and just focus on average price sensitivity or the average slope of the demand curve. But what if we start with the observation that people may differ not only in their health but also in their price sensitivity of demand? Then you have to think about who’s going to select these high-deductible plans: Are they going to be the individuals who are more or less price sensitive? That gets to this notion of what we call selection on moral hazard. This is the focus of another paper with my collaborators Liran Einav, Steve Ryan, Paul Schrimpf, and Mark Cullen (Einav et al. 2013). Here, we thought about a model in which there are three reasons that people might demand health insurance. The first is the traditional adverse selection channel in which Q

31

Q

MORAL HAZARD IN HEALTH INSURANCE

people have private information about their risk type (in this case, their health). The second is what we’ve called selection on moral hazard: The person knows that he has a high price sensitivity of demand and thinks “If you make health care cheap, I’m more likely to use a lot of it,” so this affects his demand for a plan with low consumer cost-sharing. The third is just risk aversion; more risk-averse people will always demand more insurance. An analogy from our collaborator Mark Cullen helped us think about the difference between traditional selection and selection on moral hazard. Think for a moment of the context of all-you-can-eat restaurants. In the context of an all-you-can-eat restaurant, traditional selection is that people with big appetites are more likely to go to all-you-can-eat restaurants. Selection on moral hazard is the idea that even if you have an average appetite, you know that when food is free on the margin—the marginal price of an additional entrée is zero at an all-you-can-eat restaurant—you’re going to consume a lot more than you usually do when things are priced a la carte. So people who tend to eat a lot more when the price of food is lower also find an all-you-can-eat restaurant appealing. Selection on moral hazard is thus selection on the slope, or on the price sensitivity of demand, rather than “traditional” selection on the intercept, or the level of demand. We looked into this using data from Alcoa that included employee health insurance options, choices, and medical claims. We found evidence of selection on moral hazard. The particular sign was that individuals who are higher moral hazard types—in other words, individuals who increased Q

32

Q

MORAL HAZARD IN HEALTH INSURANCE

their medical consumption more when it was subsidized more—are more likely to choose more coverage. Thus, not only do sicker individuals seek more medical coverage (traditional adverse selection), but also individuals with a higher price sensitivity of demand seek more coverage (selection on moral hazard). Quantitatively, in our particular setting, selection on moral hazard turned out to be quite important. It was almost as important as traditional adverse selection, at least in our particular context, in determining people’s plan choices. So why do we care? Again, let’s remember the goal, which is to translate experimental estimates of the impact on spending from randomly assigning people to different plans to a forecast, for example, of what happens when we offer these tax-subsidized high-deductible accounts. The key point is in most real-world settings, including that of health savings accounts, people get to choose whether to be in a highdeductible plan or not—we don’t randomly assign people to the plan. The traditional approach to forecasting how health spending is going to respond to the introduction of high-deductible plans is to estimate what percentage of the population will take up these plans given how they’re priced, and then applying the average estimate of moral hazard that we get from experimental variation, such as the experimental variation in the RAND experiment. Remember that that experimental variation was key for being able to get the causal effect of one’s health insurance plan on one’s medical spending. But here, the very feature that’s solving the causal inference problem, namely random assignment, is shutting down the possibility of selection, Q

33

Q

MORAL HAZARD IN HEALTH INSURANCE

which can be very important for the spending effects in a context where people can choose their plans. In particular, consider for example that we are trying to predict the spending reduction associated with offering a $3,000 per family deductible plan. If we take the traditional approach and assume that the people who are buying those plans are random with respect to their moral hazard type, we would use our average moral hazard estimate and predict a decrease of about $350 in spending per employee if we go from the current plan to a high-deductible plan. However, if we recognize based on our estimates that the people who are going to select the high-deductible plan with less coverage are precisely the least price sensitive people— in other words, the low moral hazard types—the spending reduction that will result by these people moving into a highdeductible plan is going to be much lower, because those who choose the high-deductible plans are less responsive than average to consumer cost-sharing. Depending on how the plan has been priced, the utilization effect (or spending decrease) can be lower by a factor of two or three. This will have very important implications for how introducing a plan when people are given a choice, as is usually done, is going to reduce health spending.

SPENDING GROWTH The third and final thing I want to discuss is really, in some sense, the key issue in health care spending: It is not the level of health care spending, but rather the growth of health care spending. Thus far, what I have discussed is the likely effect Q

34

Q

MORAL HAZARD IN HEALTH INSURANCE

of introducing a high-deductible plan on the level of spending. I have suggested that we may get much less of a spending reduction than we would have conventionally assumed, both because people are forward looking in their medical consumption decisions, so they don’t respond fully to just the current deductible, and because of the issue of selection on moral hazard, where people who choose the high-deductible plan may be those who respond least to the deductible. But the real story in the health care sector is not the level of spending so much as it is growth, especially relative to GDP. Figure 3 shows the growth in health expenditure as a share of GDP. Health expenditure as a share of GDP has grown from about 5 percent in 1960 to about 18 percent in 2010.

20

Percentage

15

10

5

0 1960

1970

1980

1990

2000

Year Figure 3 National health expenditures as a percentage of GDP, 1960–2010. Source: Centers for Medicare and Medicaid Services (2012). Q

35

Q

2010

MORAL HAZARD IN HEALTH INSURANCE

What is behind the growth of health spending? This is an area of rare consensus among economists. Joseph Newhouse and others have done work in this area, looking at the roles of, for example, the aging of the population or other demographic factors (Newhouse 1992). While these all play some role, there is widespread consensus that technological change in medicine is the driving force behind the growth in health spending. But this just kicks the can down the road: What then drives technological change in medicine? The particular question I’m interested in is this: What role does health insurance play in affecting technological change in medicine? Another way of putting it is to note that, thus far, all of the analyses I have been describing have asked: How does my health insurance affect my medical spending, assuming that nothing else around me changes? The health care system is held constant in these analyses; this is what we call partial equilibrium analysis. In fact, however, major expansions of health insurance coverage incentives to develop and adopt new technologies have general equilibrium effects on the health care sector that are very different, perhaps bigger or perhaps smaller, than these partial equilibrium effects we’ve been describing so far. This is something I studied in an earlier paper looking at the introduction of Medicare in 1965 (Finkelstein 2007). Prior to the current health care reform, the introduction of Medicare was the single largest expansion of health insurance coverage. It provided health insurance to virtually all Americans aged 65 and older. To give you a sense of magnitude, prior to Medicare about three-quarters of the elderly Q

36

Q

MORAL HAZARD IN HEALTH INSURANCE

were uninsured. And the elderly were about 10 percent of the population. Medicare therefore provided insurance coverage where there was none before to about 7.5 percent of the U.S. population, which is roughly similar to the share of the uninsured in Massachusetts prior to the 2006 Massachusetts reform. The Affordable Care Act of 2010 could cover about 11 percent of the population, so the expansion is roughly similar in magnitude. To estimate the impact of introducing Medicare, I used the fact that prior to Medicare, rates of health insurance coverage were very different across different regions of the country. On average, about a quarter of the elderly had private health insurance coverage prior to Medicare, but the rates varied a lot across regions. So for example, if we look at the percentage of the elderly who didn’t have insurance, who would be newly insured by Medicare, in New England Medicare increased the fraction with insurance by about 50 percent, but in the East South-Central United States Medicare increased the fraction with insurance by about 90 percent. I looked before and after the introduction of Medicare in areas where Medicare had more or less of an impact on insurance coverage and looked at differential changes in the growth of health spending, and I found enormous spending effects. By 1970, only 5 years after Medicare was introduced, I estimate that hospital spending was almost 40 percent higher than it would have been absent the introduction of Medicare. That’s total hospital spending, not just hospital spending on the elderly. One back-of-the-envelope extrapolation result based on this is that the spread of all insurance, Q

37

Q

MORAL HAZARD IN HEALTH INSURANCE

public and private, which occurred between 1950 and 1990 may be able to explain half of the sixfold growth in the real per capita spending over this time period. These results suggest that the spread of insurance played a very big role in driving health care spending growth over the second half of the twentieth century. This is contrary to what would be obtained if using, say, the RAND estimates, which are partial equilibrium estimates and suggest that the spread of health insurance explains maybe only a tenth of the growth of health care spending over that period (Newhouse 1992). What is the difference? One difference is that RAND is a real randomized experiment. One can be pretty sure of those numbers, whereas I had to do my best to compare these different areas of the United States and control for different trends. There’s always a possibility that that’s part of what’s going on, but I actually think—and what I have evidence for in the paper—the analysis of the introduction of Medicare is capturing general equilibrium (i.e., systemwide) effects that the RAND Health Insurance Experiment cannot. In RAND, the sample is six thousand people across the United States, so we are getting the effect of someone newly insured on his or her health care use, holding constant the health care environment: the doctor the individual sees and the hospital the individual goes to are not doing anything different because a few more people have health insurance. However, when we suddenly have 7.5 percent of the population newly covered—and it is the highest-spending part of the population (the elderly)—it is increasing aggregate demand for health care. This can Q

38

Q

MORAL HAZARD IN HEALTH INSURANCE

provide incentives to providers to change the overall practice of medicine. In fact, I find evidence that the introduction of Medicare encouraged the adoption of new medical technologies. As I noted earlier, there’s widespread consensus that technological change in medicine is behind the growth in health care spending. Now we find that when large-scale insurance changes lead to a big aggregate increase in demand, hospitals have an incentive to adopt new medical technologies. People will use these new technologies because they are not paying for them out-of-pocket, and they presumably would not have used them as much when they had to pay for them out-of-pocket. We can trace that line of reasoning back further, not just to how insurance affects the adoption of new medical technologies, but also to how it affects the development of those technologies in the first instance. This is harder to do in the case of Medicare, because it is a national or global innovation market, but from other work that I’ve done (Finkelstein 2004) and that Daron Acemoglu and Josh Linn have done (Acemoglu and Linn 2004) looking at pharmaceutical innovation, we have found that when one increases the expected size of the market for a particular drug or a vaccine by requiring that people get it, subsidizing it or by having a growing demographic base for it, one sees increased new clinical trials and increased new drug approvals for these drugs and vaccines. It therefore looks like insurance, by increasing demand because it lowers the price of medical care, encourages both the adoption of new medical technologies—as we saw in Q

39

Q

MORAL HAZARD IN HEALTH INSURANCE

the case of Medicare—and, further down the pipeline, the innovation and development of these new technologies. Another way of saying this is that we usually think of insurance as providing insurance against current expenditure risk, but part of what insurance may be doing is changing the nature of expenditure risk that we will have to insure in subsequent generations.

CONCLUSIONS AND FURTHER THOUGHTS In 1963, Ken Arrow proposed the concept of moral hazard in health insurance, the idea that health insurance may increase the demand for medical care. That creates a fundamental tension for health policy that is trying both to cover the uninsured and simultaneously reduce the level and growth of health spending. The challenge posed in Arrow’s paper to subsequent generations of economists was whether we could verify that this theoretical notion of moral hazard and health insurance actually existed, and whether we could quantify its magnitude and explore its nature and implications. How have we risen to this challenge? There is compelling evidence from randomized trials that health insurance affects medical spending. Those who say otherwise are ignoring the evidence at their own peril. This is a fact of life. It may not be what we wished for, but we have to think about it, grapple with it, and think about its implications for policy. Some of the implications have to do with what price the individual considers: is it the spot price? the end-of-year Q

40

Q

MORAL HAZARD IN HEALTH INSURANCE

price? I discussed this in the context of high-deductible health insurance accounts, but one can also think of this in terms of Medicare Part D, the prescription drug program for the elderly. One can consider the famous “doughnut hole,” which is a range of spending in which suddenly an individual’s price goes back up to 100 percent. How the elderly respond to a policy with a doughnut hole depends on whether they are thinking about it at the beginning of the year or not. Additionally, issues of selection on an individual’s moral hazard type and how health insurance may affect the development of new technologies are important for the full “moral hazard” analysis. I think we have made some progress, but, naturally, important questions remain. Let just mention a few of them, although I am sure there are many more: Where does moral hazard come from? How much of it comes from the patient and how much of it comes from the doctor? Clearly, the decision to go to the doctor initially is primarily driven by the patient, but once there, how much of the decision of what care to get and how to adjust care based on the patient’s insurance coverage is driven by the patient and how much by the doctor? This has implications for designing contracts to reduce excess moral hazard. Another key issue is the welfare cost of moral hazard. The price of medical care does not reflect its true marginal social cost. As just one example, consider prescription drugs, which due to the patent system have prices that are much higher than their marginal social cost of production, which is approximately zero. It is not obvious what is the welfare cost of health insurance inducing increased consumption of Q

41

Q

MORAL HAZARD IN HEALTH INSURANCE

an already developed product; it may be that the real welfare consequences are via the effect of health insurance on the incentive to develop a product in the first place. These are important issues to think more about.

REFERENCES Acemoglu, Daron, and Joshua Linn. 2004. “Market Size and Innovation: Theory and Evidence from the Pharmaceutical Industry.” Quarterly Journal of Economics 119(3):1049–1090. Akerlof, George. 1970. “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism.” Quarterly Journal of Economics 84:488–500. Aron-Dine, Aviva, Liran Einav, and Amy Finkelstein. 2013. “The RAND Health Insurance Experiment, Three Decades Later.” Journal of Economic Perspectives 27(1):197–222. Aron-Dine, Aviva, Liran Einav, Amy Finkelstein, and Mark Cullen, 2012. “Moral Hazard in Health Insurance: How Important is Forward Looking Behavior?” NBER Working Paper No. 17802. National Bureau of Economic Research. Arrow, Kenneth. 1963. “Uncertainty and the Welfare Economics of Medical Care.” American Economic Review 53(5):941–73. Centers for Medicare and Medicaid Services. 2012. National Health Expenditure Data. Available at: http://www.cms.gov/ResearchStatistics-Data-and-Systems/Statistics-Trends-and-Reports/ NationalHealthExpendData/NationalHealthAccountsHistorical. html. Congressional Budget Office. 2012. Long Term Budget Outlook. Available at: http://www.cbo.gov/sites/default/files/cbofiles/ attachments/06–05-Long-Term_Budget_Outlook.pdf. Einav, Liran, Amy Finkelstein, Stephen Ryan, Paul Schrimpf, and Mark Cullen. 2013. “Selection on Moral Hazard in Health Insurance.” American Economic Review 103(1). Q

42

Q

MORAL HAZARD IN HEALTH INSURANCE

Ehrlich, Isaac, and Gary Becker. 1972. “Market Insurance, SelfInsurance and Self-Protection.” Journal of Political Economy 80(4):623–48. Finkelstein, Amy. 2004. “Static and Dynamic Effects of Health Policy: Evidence from the Vaccine Industry.” Quarterly Journal of Economics 119(2):527–64. Finkelstein, Amy. 2007. “The Aggregate Effects of Health Insurance: Evidence from the Introduction of Medicare.” Quarterly Journal of Economics 122(3):1–37. Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joe Newhouse, Heidi Allen, Katherine Baicker, and the Oregon Health Study Group. 2012. “The Oregon Health Insurance Experiment: Evidence from the First Year.” Quarterly Journal of Economics 127(3):1057–1106. Gladwell, Malcolm. 2005. “The Moral-Hazard Myth.” New Yorker, August 29. Newhouse, Joseph P. 1992. “Medical Care Costs: How Much Welfare Loss?” Journal of Economic Perspectives 6(3):3–21. Newhouse, Joseph P., and the Insurance Experiment Group. 1993. Free for All: Lessons from the Health Insurance Experiment. Cambridge, Mass.: Harvard University Press. Pauly, Mark. 1968. “The Economics of Moral Hazard: Comment.” American Economic Review 58(3, part 1):531–37. Rothschild, Michael, and Joseph Stiglitz. 1976. “Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information.” Quarterly Journal of Economics 90:630–49.

Q

43

Q

COMMENTARY JONATHAN GRUBER

I

was thrilled to act as a commentator on Amy Finkelstein’s lecture in honor of Ken Arrow. Amy is a colleague of mine at MIT, and I remember diligently working to recruit her to come to graduate school there when she was on the Council of Economic Advisors in the late 1990s. I would like to think that Amy’s MIT education is responsible for some of her enormous success since that time—but to be honest, we’re probably just riding on her coattails. And these coattails are indeed impressive. Amy is now one of the best health economists in the country as well as one of the best overall applied microeconomists in the country. It’s not just me that thinks so; just two weeks after giving her lecture, Amy was awarded the John Bates Clark Medal by the American Economic Association. In dozens of innovative articles, some of which are mentioned in the previous chapter, Amy has carried on a great MIT tradition—which started with Robert Solow and Paul Samuelson and continues through Paul Joskow and many

COMMENTARY

others in our department—of combining rich theoretical insights with compelling empirical work. After many years of watching Amy’s career skyrocket, in 2012 I was fortunate enough to coauthor a project with her—the study of the randomized health insurance trial in Oregon that Amy writes about in this book. I have to use the term coauthor quite loosely—you could imagine Amy as a rocket, zooming off into the stratosphere, and me just trying to hang on. I received literally dozens of e-mails from Amy each day, so fast that I couldn’t respond before I got another one, and eventually I basically just dropped out. She was kind enough to continue to keep my name on the paper, but I’m still shuffling through the e-mails years later. It’s really just a pleasure and an enormous privilege to get to work with Amy on a project. For her chapter in this book, Amy focused on one of the key health information asymmetries, first described by Arrow in his 1963 article “Uncertainty and the Welfare Economics of Medical Care.” It’s a testament to how rich Ken Arrow’s work is that when considering this article, the first thought that comes to mind is not the moral hazard contribution, but rather the adverse selection contribution and the notion of insurance market failure through adverse selection. And it’s further testament to Amy’s own breadth of expertise that she could have presented an entire lecture on the adverse selection issues. She has contributed much to this topic in the past couple of decades, including her work with Kathleen McGarry. That she could actually take on a second key element of Ken’s article and talk about moral hazard is impressive. Q

46

Q

COMMENTARY

I would just like to make four points complementary to what Amy described, coming back especially to what she touched on at the end of her chapter. The first point is that she only focused on one kind of moral hazard, what she called the ex post moral hazard, although she also mentioned ex ante moral hazard, or the notion that people might not take care of themselves because they are insured. I think ex ante moral hazard is one of those things that economists like to spend a lot of time on but does not matter very much. We actually did investigate ex ante moral hazard in the Oregon Health Insurance Experiment; we looked at whether people would, for example, smoke more or exercise less when they had health insurance. So far, we do not have the evidence that that is the case. But there is another kind of moral hazard that Amy touched on just briefly at the end of her chapter, which I think may be even more important than the one she discussed. This is provider-side moral hazard—which, to be clear, is distinct from the general equilibrium effects Amy described in her Medicare work. This issue is best summarized in the saying that having a doctor tell you how much medical care to get is kind of like having a butcher tell you how much red meat to eat. What we face in the United States is a broken fee-for-service health care system where physicians and providers are paid based on how much care they deliver, not on how healthy they make you. This is the focus of most discussions of health care cost control today. While many of us are fascinated by individual-side moral hazard, in fact most of the discussion that dominates policy circles is not about that; it is about the Q

47

Q

COMMENTARY

provider-side moral hazard and how we can move from our existing fee-for-service system toward alternative systems; for example, accountable care organizations, where providers are brought together under one global reimbursement system and are reimbursed some aggregate amount based on enrolling patients, rather than treating them. There is an enormous amount of evidence that providerside moral hazard matters. I think the most compelling example was from a major change in Medicare in 1983, which was probably the biggest change since the program was introduced in 1965. Medicare led the way in moving from a fee-for-service reimbursement system to a prospective reimbursement system, where the program paid hospitals not on the basis of actual treatment but on the basis of diagnoses. When Medicare did so, there was an enormous change almost overnight. The average length of stay of the elderly at hospitals fell by 15 percent within 1 year. And yet there was no effect on elderly health. Basically, there was excess care that was not improving the health of the elderly, and getting rid of that saved a lot of money and lowered utilizations. The provider-side consideration is an important part of moral hazard. This is not to dispute what Amy said, but to add to a critical issue she described at the end of her chapter, which is the complementarities between the patientside moral hazard and the provider-side moral hazard. These communities are talking past each other: If we talk to the provider-side moral hazard camp, they say, “Oh, that patient-side stuff is not important, it doesn’t really matter. Co-pay is deductible, that’s all just regressive and unimportant. We’ve just got to focus on how we reimburse providers.” Q

48

Q

COMMENTARY

And then we have the other camp—I’m more on this side— saying “It’s really critical that we get consumers to be more cost-sensitive and to pay attention to the cost of medical care.” Both sides are clearly right, and what neither side has really tackled is the question of how these two types of moral hazard interact. In particular, how do we make the provider-side incentives and the patient-side incentives interact? How do we, for example, set up patients in plans that charge them more if they go to providers that are on fee-for-service rather than on capitated reimbursement? How can these two things be combined? This will be a crucial issue to think about going forward. The second point concerns the interpretation of the evidence from the RAND Health Insurance Experiment and the Oregon Health Insurance Experiment. As Amy mentioned, these studies do not measure moral hazard directly; rather, they measure the effect of price on medical care utilization. And there are really two components to this measure: one of them, the substitution effect, is a moral hazard response, while the other is something economists call an income or liquidity effect. More generally, think of the fact that on the one hand, the price of medical care is changing. On the other hand, when someone has insurance, that individual has just been given a gift of money should he or she get sick, and the individual may choose to spend some of that money on his or her treatment. This is what we call the liquidity effect. That is probably what is partly going on when people get health insurance and then spend more. People would always like to spend on medical care, but they are constrained from doing so. Q

49

Q

COMMENTARY

By giving them money when they are sick, we allow them to afford that expensive operation they could not otherwise. This is not a distortion of the system; it is just getting rid of the problem of the liquidity constraints that people face. The problem? There are very different policy implications, depending on whether changed medical spending habits are due to true moral hazard or to liquidity effects, and we don’t really have great evidence to separate them—although we know the ideal experiment to run. Take two people: to one we give health insurance, and to the other we just give a check should he get sick and allow him to use it however he likes. If it is a substitution effect, they will behave differently, and moral hazard is at work. If it is just a liquidity effect, they will behave the same. I do not think we will ever get to run that randomized trial, but I’m sure Amy is clever enough to figure out a way to tease out that test in the real world. And in fact, in the context of unemployment insurance, Raj Chetty at Harvard and his colleagues found empirical and quasi-experimental ways to measure this and have shown that a lot of what we thought was moral hazard in unemployment insurance is actually just liquidity effects (Card et al 2007, Chetty 2008). As the third point, it seems pretty obvious from what Amy describes about moral hazard effects that we ought to have patient-side cost-sharing and to have patients bear some of the consequences of medical decisions. There is of course a flip side to patient-side cost-sharing, which is affordability concerns. This is why so many people are opposed to having patients face cost-sharing for their health care: They are worried that people cannot afford Q

50

Q

COMMENTARY

these costs. The reason we have out-of-pocket limits is to make sure health care is affordable. Take the example of someone who is at the poverty line, with an income of $13,000 a year. That individual cannot afford a $2,000 deductible. He or she probably cannot afford a $1,000 deductible. How can you have patient-side cost-sharing in the face of affordability issues? Years ago, Martin Feldstein proposed a plan for this (perhaps his only progressive policy proposal), which he termed major risk insurance. In his major risk insurance plan, people would pay 50 percent of the cost of their health care until they spent 10 percent of their income. Basically, people would be protected from being bankrupted by medical costs but would be price-sensitive on the margin. Whether that is a good idea or not actually depends on another aspect—income heterogeneity—which has not really been studied. In the kind of moral hazard Amy is talking about, whether a plan is good or not depends on whether poor people are more or less sensitive. If they are extremely price-sensitive, there might be a higher welfare cost, and we might want to not quite protect them as much as we otherwise would. We have to understand not just the overall moral hazard effects that Amy mentioned, but also how that affects differences on the income distribution. As the fourth point, the trade-off between setting the right marginal incentives for consuming health care and mitigating uncertainty represents another difficulty in translating Amy’s work to policy. Economists have always said that the right way to set up a health insurance plan is to have what we call co-insurance, which is to say that one would pay, say, Q

51

Q

COMMENTARY

20 percent of one’s bill up to some maximum. But people in the real world hate co-insurance. They like co-payments, and when asked why, they say because otherwise there’s too much uncertainty about what they’ll pay when they go to the doctor. Part of the problem is the lack of price transparency. We do not know what 20 percent of the doctor’s bill is going to be because it is not posted. And the other part of it is true uncertainty, as we do not know what that doctor is going to do, which tests the doctor is going to order, and so forth. So subjecting people to co-insurance subjects them to some uncertainty. We know from the pioneering work of Ken Arrow and others that people do not like this uncertainty and that they would pay to insure against it. That actually raises an interesting question of the right way to incentivize people. On the one hand, when patients face the price on the margin, they will be more deliberative about getting expensive treatments. And on the other hand is the uncertainty patients face about what they’re going to have to pay out-of-pocket. A really interesting question— which is straight along the line of Amy’s research agenda—is to understand those trade-offs between co-insurance, which economists find appealing, and co-payments, which people in the real world find appealing. How we actually resolve that trade-off going forward, I think, is significant. These are just a few ideas. Amy’s work is so excellent that to truly criticize it is impossible, so I thought I would instead offer a few suggestions as to where she could go next. I think Amy is really perfectly positioned to answer some of the questions that I’ve raised here. Q

52

Q

COMMENTARY

REFERENCES Card, David, Raj Chetty, and Andrea Weber. 2007. “Cash-onHand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market.” Quarterly Journal of Economics 122(4):1511–60. Chetty, Raj. 2008. “Moral Hazard versus Liquidity and Optimal Unemployment Insurance.” Journal of Political Economy 116(2):173–234. Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joe Newhouse, Heidi Allen, Katherine Baicker, and the Oregon Health Study Group. 2012. “The Oregon Health Insurance Experiment: Evidence from the First Year.” Quarterly Journal of Economics 127(3):1057–1106.

Q

53

Q

COMMENTARY KENNETH J. ARROW

I

was very happy to return once again to my graduate alma mater to deliver some remarks in response to Amy Finkelstein’s lecture. Columbia University today is a very different world, in many ways, than it was when I was a student. What’s taught in the economics departments and the issues of major concern have evolved—although the Great Depression does seem to have a rather contemporary echo in the recent crisis. Some things change, and some things go around in cycles. Amy’s wonderful lecture and corresponding essay for this volume drew together evidence and showed the ability to develop experiments and to exploit natural experiments with a creative emphasis on the theoretical understanding. Amy’s presentation was a great pleasure to all of us who witnessed it. We have read the papers before, but her work puts everything into a perspective that is already making a contribution to our study of our health problems. Let me take a tack that was referred to by Amy and was discussed somewhat more at length by Jonathan Gruber,

COMMENTARY

which I will call the non-economic side. In fact, part of this is in my original paper (which, I’ll note, is included as an appendix in this book). What I meant by moral hazard is basically an asymmetry of information or knowledge between two parties. (Moral hazard is just one kind of asymmetry.) I came to think about it in the context of health when I was approached by Victor Fuchs about a new project. Victor was then an officer for the Ford Foundation, which at that time was a very large grantor to research in the social sciences. Victor is one of the most distinguished health economists in the country and the world—I had met him, but we were not close. He called me with the following project, which he had gotten approval on from Lloyd Reynolds of Yale, the director of Economics and Research. Victor wanted to take three issues that involved public and private policy and have each one studied by two people. One person, who had already worked in the field, would give some kind of summary, and the second person would preferably be a theorist who had not worked in the field at all and who would see what theory could do. He wanted me to be the theorist for health. I have a tendency, which has worked out quite well for me in many ways, to take a dare. This is probably the best example, although Social Choice and Individual Values actually arose to some extent in the same way. When I am challenged to do something that I am not working on or continuing research on, I think, “Well, this is more exciting, I’ll learn something.” I thought, after all, that health is a serious issue, and if theory is any good at all—a point that can Q

56

Q

COMMENTARY

be debated—then this would be a good place to apply it and see what we have to say. Of course, apart from social choice, my big interest had been the economics of uncertainty and risk bearing. So I came at the topic trying to fit it into a general equilibrium with uncertainty concepts. I very diligently read the various books out there and the other literature that existed at the time. There were health economists even then, such as C. Rufus Rorem, Harold Somers, and Anne Somers. I read a lot of descriptive material. In insurance at that point, there was a fair amount of hospitalization insurance. It was just the beginning of insurance for what was called major medical risks, by which is meant risks outside of the hospitalization area. HMOs and things like that did not exist at the time. There were a few clinic groups, but they were a very small part of the story. These were all incipient, but the health problem looked pretty big at the time, although it looks trivial by modern standards. It was probably of the order 6 percent of national income. It was a significant problem, but not overwhelming. It seemed pretty obvious that if you have risks, you ought to have insurance. Why can’t we have much more extensive health insurance? I began trying to fill in this question. I thought it might be difficult for the insurance companies to know exactly what they are insuring against. For example, with automobile insurance, when you have a smashed windshield, there is no choice involved: you have to replace it. Even then there are problems of moral hazard, but they are not the dominant concern. If the procedure being done is preassigned, then there is no moral hazard. The diagnosis Q

57

Q

COMMENTARY

is a broken windshield, and the procedure is to replace it. But when you are sick, your diagnosis might turn out not to be what you thought it was, and it may be more expensive or less expensive. There are a lot of decisions that have to be made along the way, some of which are just questions of comfort, and some of which are really medically significant. The insurer is not in a good position to monitor what is going on. Then, I began to realize there was another problem: the relationship between the physician and the patient. How does the patient know the physician is putting forth full effort on a case? Unfortunately, I know some people who have had serious consequences from that, so it is not imaginary. I began to realize that these were the same issues; the light bulb went off, just like in the cartoon strips. I had spent several months trying to do a diligent job reading all this material, and it was almost workman-like, where I said how each aspect was related to theory. Suddenly, I got the idea that there was a general principle here, which was that in these relations there was asymmetric information. A doctor’s function in life, by definition, is to know things you do not know. The insurance company is also at the mercy of the physician and patient. Of course that raises an issue with relation to the concepts of moral hazard and adverse selection. Now, I had had some slight exposure to the actuarial world. Namely, at one point in my life, not knowing what I was going to do for a living when money had become a problem and my family was badly hit by depression, I was concerned with some secure occupation. Academic life was sort of another world Q

58

Q

COMMENTARY

that I did not know about, so I was thinking of becoming an actuary. In reading about it, I had come across words such as moral hazard and adverse selection. I would not have thought of them as part of my working knowledge, but somehow, though I had read about them probably 10 or 20 years earlier, those concepts came to me suddenly. I realized, “Oh, this is what the insurance company people are always talking about.” Now the question is, how do we deal with asymmetric information? This is a defect. In other words, we have a system of ideal future markets with conditional uncertainties. But if there is asymmetric information, there’s a problem. You and I cannot make a contract conditional on something if only I know whether it happened or not. We have an immediate explanation of market failure. What does this create? If we have prices for everything, ethical considerations do not really enter the story, but as soon as we have this asymmetric information and the failure of prices, we have ethical problems, which a true market system does not have. There is some kind of a social benefit to ensuring truthtelling. Of course there is a vast literature on mechanism design. Personally—though this is a very important literature and it has its role—I think there are all sorts of reasons why it is not an adequate substitute for market failure due to asymmetric information. In some cases, it will be. Even in the financial world, we have seen so much of this, because it is very difficult to devise incentives. In fact, we have some evidence that the incentives supplied have not functioned as well as might be desired, if I put it delicately, and it is not clear to me there is any way of completely correcting that. Q

59

Q

COMMENTARY

One of the points that Amy raised briefly and that Jon wrote about a little more extensively was that if you look at the system, the decision to buy and to use medical care is by no means a decision of the patient only. I rely on the doctor just as I rely on an attorney. There is a reason why professions have a special role. They are marked by strong inequalities in knowledge; basically what a doctor’s services supply is not skill—though that matters—but knowledge. We can argue about why there exists specialization, where a person is acquiring knowledge that can be used over and over again. Socially, it pays for some people to have knowledge and use that knowledge repeatedly with new patients or clients, rather than everybody being educated in the medical field. That creates, automatically, an asymmetry of information in the whole system, including physicians, hospitals, and HMOs. When insurers started gradually going into the health business, they started checking on the doctors. The doctors were recommending treatments, and the insurance companies were rejecting the recommendations—an issue that’s now become notorious. The insurance companies or HMOs are using their knowledge, too. The result is a very complicated kind of information system in which the patient’s role becomes very minimal. This worries me, particularly the most striking question of the differences in the moral hazard response. What exactly is the source of that? Is it that one is getting different doctors, for example, or different localities with different conditions? Medical ethics has always been considered a very important matter. There was a whole school of thought, especially around the University of Chicago, arguing that restrictions Q

60

Q

COMMENTARY

on medical procedures and medicines, and all these rules and ethics, are essentially devices for improving monopoly power. Milton Friedman was a great advocate of that point of view. He went so far as to argue that we should not even license doctors and that we should have free entry into the medical profession. But I think this misses the point very considerably. We find, for example, medical associations putting out information designed to prevent one from getting ill, which presumably could hardly be regarded as a profit-maximization procedure. In my original article, I instance the fact that doctors and hospitals do not advertise as an example to signal that they are not maximizing profit, although unfortunately now that particular story is quite wrong. When I pick up the papers and I see very respectable hospitals, such as Memorial Sloan Kettering, explaining why they are the best place to go for cancer or whatnot, I am still a little shocked, but I suppose younger people are not. We rely on medical ethics, and we have since Hippocrates. Similar things happen in other professions. Let me mention some evidence. I am not an expert, and some of these facts—well, some of these alleged facts—may not be facts. With the introduction of new medications and devices, there is always the matter of who decides to use them. One of my students was studying that, and what she found was that it was very hard to locate who made these decisions. Every HMO would say, “It’s up to the doctor, and if the doctor thinks it’s necessary, we buy it.” But it is clear that some HMOs are very resistant to that. This is a subtle thing that Q

61

Q

COMMENTARY

is not caught by the market system. As studies have claimed and as the Dartmouth group has emphasized, medical costs vary greatly by region, with apparently no particular relation to outcomes. If you take international comparisons, we have enormous differences in medical costs. Before and after it was pretty much the same allocation problem; it is not at all clear why there should be this difference. After all, Canada does not face different issues than the United States, so why is health care in Canada so much cheaper? As I said, if insurance raises prices, it should have much better care coverage than we do. One would expect to think it worked the opposite way, and one would expect Canada to be more expensive than the United States. If we look at the French or Israeli system, all these seem to involve giving much better coverage and yet somehow getting by with much lower costs. Nobody is even close to the United States. It is widely held, and I believe probably correctly, that if you take different HMOs, their costs are quite different. Locally in California, everybody says Kaiser gives much better value than others, and yet Kaiser has not spread in spite of efforts, which is interesting. The workings of the market should imply that Kaiser and Kaiser-like plans should spread, but they have not spread nearly as rapidly as what one might expect. Economists always hate to invoke the concept of culture. It has been around at least since Max Weber’s identification of emergence of capitalism with religion in 1905. In particular, if we go to our organizational behavior colleagues at

Q

62

Q

COMMENTARY

business schools, we find they are always talking about corporate cultures. If we have an organization with people working together, they develop ways of working that are hard to change. Newcomers come in, and they are going to be adapted to the culture rather than thinking of ways to change it. The fact that other firms are doing better does not necessarily mean that in the long run, the inefficient firms will be eliminated— especially in the medical field, where competition for various reasons is not so severe. We certainly see differences even in firms that are much more exposed to the market. We clearly need to develop a culture of accountability. Obviously there is room for accountability and accountable care organizations, which have a culture of cost-effectiveness. How this is going to be achieved, I must say, is a mystery. Playing with the contract reimbursement systems and going to capitation instead of fee-for-services matter a great deal, but a lot more has to be achieved by organizational claims. Obviously, as organizations perpetuate themselves, they need some kind of outside leadership, but not, I think, central control. Something like the British National Institute of Clinical Excellence, which gives statements about the value of new drugs with some idea of their costs, could play some important role. Now, how this is going to be studied experimentally I’m afraid I must leave to those much more qualified than myself.

Q

63

Q

COMMENTARY JOSEPH E. STIGLITZ

I

t is a real pleasure for me to once again serve as a discussant for the Ken Arrow Lecture. Amy has provided an insightful lecture building off one of Ken’s seminal contributions. As a graduate student, I was inspired by his 1963 paper1 and by his Johansson Lecture, “Aspects of the Theory of Risk Bearing,2” and it provided much of the motivation for my work with Michael Rothschild on the theory of insurance with imperfect information.3 Let me begin by addressing the most fundamental issue in which we are all interested: the design of a health care system, including an insurance system that would increase our welfare. When we think about this, one of the central problems in the analysis is that in most areas of economics, we begin with the presumption that markets are efficient, but this is an area where we should begin with the presumption that markets are not efficient. That really changes the analysis in crucial ways. One of the reasons that markets are not efficient has to do with the very general theorem that Bruce Greenwald and

COMMENTARY

I proved a number of years ago, namely that whenever there are information asymmetries, whether of the adverse selection or moral hazard kind, or incomplete risk markets, then the economy is not efficient.4 (Virtually all markets are characterized by information asymmetries, but such information imperfections are particularly important in the health care market.) It is not constrained Pareto efficient, taking into account the cost of information and the cost of setting up markets. There are always interventions that could make some individuals better off without making others worse off. The doctrines that you hear on one side of the political spectrum—that if we just left it to the markets everything would be right—have no intellectual foundation, at least in the sphere of health economics. There is a need for government, and that is why the issue of how we design an appropriate set of institutions is absolutely key. The second observation I want to make really echoes what Ken Arrow stated: Our current system is not working. To me, the primary issue is not that we are spending so much on health care. I do not have a view about the right amount that we ought to be spending on health care. People like to live longer (a preference which seems particularly evident as people get on in years). It is not that we spend 17 percent of our GDP on health care that is the problem; it is that we get so little for how much we spend. The real problem is the inefficiency of our health care system. The fact is, as Ken pointed out, that we spend much more than any other advanced industrial country both on a per capita basis and as a percentage of our GDP, yet our health statistics are worse than those of other advanced industrial countries. In fact, in Q

66

Q

COMMENTARY

some dimensions (for instance, in infant mortality), health statistics, especially in some parts of the country, are comparable to (or worse than) those of some developing countries. This should be an embarrassment to us as a country. Some of that has to do with the first issue that Amy described, the problem of access. Large fractions of America do not get access to adequate health care. I am going to come back to that in a second. One of the questions that we ought to be focusing on, which is a little bit outside of Amy’s chapter itself but which is central to the welfare and policy analysis, is, why is our health care system so inefficient? The point that Ken raised is absolutely right: Americans are not basically less healthy than those in other countries. It is not as if only sick people immigrated to America. In fact historically, we always talk about the selection effect going the other way: it was not the sick, the ill, and the genetically defective that came here. It is not as if the weather here on average is that much worse than the weather in other places; there are places that are better, I know, but it’s not the worst. There is no inherent reason why we should spend so much to get so little. It is also clear, as Ken pointed out, that there are information asymmetries on both sides of the Atlantic and the Pacific, so it is not as if this is a uniquely American problem. Moral hazard, adverse selection, and problems of monitoring and motivating doctors exist everywhere. Evidently, the way we have solved those problems is not as good in some respects as the ways others have. Part of the reason has to do with old-fashioned rent seeking. We have created a health care system rife with institutions Q

67

Q

COMMENTARY

that are very good at rent-seeking. We have institutions that do a better job at rent seeking than those in other countries. We have a public sector that has less control, so it is not a surprise that we wind up having an expensive health care system. Emblematic of that is the provision in the 2003 Medicare Drug Benefit that does not allow the U.S. government, the largest purchaser of drugs, to negotiate with the drug companies. That provision is estimated to cost in excess of $50 billion a year. That is just a single instance, but there are others. The United States had a Medicare program called Medicare Advantage. Medicare Advantage was basically identical to Medicare, but it was in the private sector and it cost about 20 percent more than the comparable public program. It did not do anything better. Why did we have it? Well, we had it because the health insurance industry lobbied to have it, and the industry made money from it. Transaction costs are a bad thing for most of us, but for the health insurance industry they are a good thing, because that is what it lives off. Ken raised another significant point: (medical) practices vary across the country and between countries. There are many practices that are not cost-effective, so in some places we observe that procedures such as tonsillectomies are expensive but have no evident benefit. One of the many good things in the Affordable Care Act was a provision for a more scientific study of these practices, which one hopes would help to create norms where we do not do costly things that do not yield any benefits. This is at least the beginning of a more scientific discussion of medicine. The final point I want to make has to do with the development of economic theory. I was very much influenced in my Q

68

Q

COMMENTARY

work by reading Ken’s Yrjö Jahnsson lecture, Aspects of the Theory of Risk-Bearing, as well as his paper on moral hazard. These were seminal works that began to help us think about problems of the economics of information asymmetries. The underlying problem arises from imperfections of information. In the case of moral hazard, we cannot observe the input, or we can observe it only imperfectly. That is a real problem. We do not know what the doctor is doing, what he should do, or what the consequences are of what he does (or does not do). Even if he did the right thing, the patient could have a disease that will have a bad outcome, so we can’t judge whether the doctor has done the right thing by whether the patient survives. He may have done the right thing and the patient dies anyway, or the patient may survive although the doctor did the wrong thing. It is really difficult to have pay-by-performance. The models that Ken, I, and others studied some 30 years ago were always very simple. We looked at a single example of asymmetry of information, a single instance of moral hazard, a single instance of adverse selection. What the discussion in this book has highlighted is that most interesting situations are multiple-information problems. The doctor does not know the illnesses of the patient, the patient does not know whether the doctor is doing the right thing, and the insurance company does not know about either of those two. In the instance just described, there are three moral hazard problems. But in the insurance market, there are additional problems of information asymmetries and imperfections. The insurance company does not know the patient’s health condition, Q

69

Q

COMMENTARY

and the patient does not know whether the insurance company is going to pay when the patient actually needs it to pay (when the patient believes that the insured-against event has occurred). It is striking to me that in the ensuing years, remarkably little theoretical research has been done trying to analyze models where these multiple-information asymmetries exist. The reason is very understandable, because the problems get very complex very quickly. One way of thinking about it is that when there are both problems of moral hazard and adverse incentives, we have both incentive compatibility and self-selection constraints, and these can interact in complex ways, making the analysis very, very difficult.5 But the problems that we face at the practical level all involve these multiple levels of information asymmetries. In one way or another, eventually we’re going to have to cut through the complexity and find out what the right approximations and simplifications are that will allow us to analyze models with multiple information asymmetries.

NOTES 1. Arrow, Kenneth J. “The economic implications of learning by doing.” The Review of Economic Studies (1962): 155–173. 2. Arrow, Kenneth J. “Aspects of the theory of risk-bearing”. Yrjö Jahnssonin Säätiö, 1965. 3. Rothschild, Michael, and Joseph Stiglitz. “Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information.” The Quarterly Journal of Economics 90.4 (1976): 629–649. It also provided much of the motivation

Q

70

Q

COMMENTARY

for my subsequent work with Richard Arnott on moral hazard. See, e.g. R. Arnott and J. E. Stiglitz, “The Basic Analytics of Moral Hazard,” with R. Arnot, Scandinavian Journal of Economics, 90(3), 1988, pp. 383–413. Reprinted in Selected Works of Joseph E. Stiglitz, Volume I: Information and Economic Analysis, Oxford: Oxford University Press, 2009, pp. 417–443. 4. Greenwald, B. and Joseph E. Stiglitz, 1986. “Externalities in Economies with Imperfect Information and Incomplete Markets.” Quarterly Journal of Economics 1(2 May): 229–264. 5. Jungyoll Yun and I have made some progress in J. E. Stiglitz and J. Yun, “Optimality and Equilibrium in a Competitive Insurance Market under Adverse Selection and Moral Hazard,” NBER Working Paper 19317, August, 2013.

Q

71

Q

DISCUSSION

A

fter the presentation of the lecture and the commentaries, audience members were given the opportunity to ask questions. Below is the resulting discussion.

It seems to me that cross-country evidence argues against moral hazard, in the sense that countries that have more comprehensive coverage have lower costs in terms of proportion of GDP spent on health. Is, then, the problem of moral hazard not with consumers of health care, but with providers of health care? Presumably, the consumers of health care have similar behavior with similar incentives, whereas what is different between the systems are the incentives for the providers of health care. Is that where the problem lies and where research should be concentrated? AUDIENCE QUESTION: There was no discussion about providerside moral hazard. Actually patients only control, in their demand, about 10 percent of all costs. Most costs are in the hands of specialists in hospitals who have multiple AUDIENCE QUESTION:

DISCUSSION

profits, gains, and incentives from ordering more tests and services for their professional well-being and their income. In your studies, did you have a control to examine the degree to which the increase in demand in cost, which you’re attributing to individuals, is accounted for by providers? AUDIENCE QUESTION: How does preventative care come into some of this modeling around increased use and potential reductions in future spending? STIGLITZ: Thank you. Can I just add one question, before Amy responds? There are two additional points I wanted to make. One of them goes to the welfare issue. Normally we think of technological advances as a good thing. There is a little bit of a hint, Amy, that you are saying that technological advances were driving costs, and as there is this paranoia about cost, there is the suggestion that it was a bad thing. Let me give you my answer. Some of the technology changes are actually good things, they are extending lifetimes and they are enhancing welfare, but in fact, the cost that we pay for them is not the marginal cost. They have large ranks associated with them, so that our system of producing technology is not efficient, and our system of using it is not efficient. For instance, in some countries more expensive drugs can be used only if it can be shown that they have a benefit. Australia maintains a list of drugs that can be used, whereas the United States does not have those kinds of control mechanisms. In a way, while technology can have potential for increasing welfare, the way we have used technology and the way our market distorts Q

74

Q

DISCUSSION

the production of technology means sometimes it is not welfare enhancing. The second point was really related to the last question. Amy, you have talked about the RAND study and a couple of the others, including the Oregon study. You talk about the usage hospital admissions now, but that does not address long-term health care costs. If some of these people are going to the hospital today when they would not have gone, they might have wound up with much worse diseases; that is to say that their problems would have accumulated. It is not obvious that more usage is going to lead to higher health care costs. FINKELSTEIN: There were really three sets of questions. One had to do with moral hazard on the provider side and the role of providers versus patients. Jon didn’t mention, but I’ll mention that he is not only my colleague at MIT, he was originally my thesis advisor at MIT. I think his discussion in some sense embodied what the ideal thesis advisor does. They encourage you, and they tell you what you’re doing is wonderful. Then as you are walking out the door, they make a side comment that basically very gently suggests you’ve entirely missed the point, but you leave feeling really pepped up and encouraged. Three days later you realize, now I understand what I need to be working on. I think his comments and the first two questions that came from the floor do really emphasize that one of the key places where we need to shine the flashlight now is in thinking about incentives on the provider side and not just the issue that I was alluding to of how much of the Q

75

Q

DISCUSSION

patient response is influenced by the physician. We need to think about how much of what the physician is doing is a response to his or her own financial incentives. That is an area where we need a lot more work. The second question that came up was on preventive care and future health care spending. We did look at this in Oregon, and the results from the first year show that use of preventative care increased, such as use of mammograms, prostate screens, checking blood cholesterol levels, et cetera. It is not clear what the long-term benefits of these are. For example, in the course of our study, the recommendations on mammograms changed, so we did not know how to code recommended preventive care— recommended as of when? The question Joe raises about the long-term future spending is something we would like to follow. We cannot follow it in the experiment, because the experiment itself lasted only 2 years, but we are getting physical measures of what happened to a patient’s blood pressure, blood cholesterol, whether the patient was taking blood pressure medication, and so on. We can use those to try to forecast what the long-run health spending would be. Jon may have a different view, but I have to say I am pessimistic that this will actually, in the long run, reduce spending. Sometimes I feel that setting that up as an objective misses the point. It is okay to spend money on your health: if health insurance increases health spending, that is not necessarily a bad thing. It would be nice if there is a free lunch, and we could actually save money

Q

76

Q

DISCUSSION

and improve people’s health at the same time. But I’m pessimistic on that. That again relates to Joe’s last point, which is a very important point to conclude on. Joe talked about technological change driving health care spending, and how that suggests that technological change is a bad thing. Of course, one can also put up a graph that shows that survival, particularly for people 65-plus, has increased dramatically over the past 40 or 50 years, and that’s in large part due to the technological change. I think Joe is absolutely right that the question is not whether we need more or less technological change, but how we align incentives to get the socially valuable type of technological change. I’m a little less sanguine than Joe about using lists of drugs, as in Australia. There are certainly some that we think almost everyone should do and some things that almost everyone should not, but there is probably a lot of heterogeneity both in the treatment benefits and the preference benefits. It’s hard to know how to work that out. ARROW: I have said to several people that the real way to control costs is to abolish the National Institutes of Health. Someone said they would save us $50 billion a year, and I said that’s not the point. It’s all these expensive medications that result from this research. Obviously we don’t believe that. There is this rather heroic calculation by Robert Hall and Charles Jones arguing that the idea that we should be spending 30 percent of our GNP on health is by no means absurd. The numbers

Q

77

Q

DISCUSSION

are sort of made up, but they are not totally incredible numbers, that’s the point. What is creating the health problem and why it is a special problem is that it is a matter of public interest and there are taxes involved. It would be less complicated if somehow we could do all this and insure privately. There is an interesting ethical dilemma here: We don’t like to see people die if the medical care is available and the only problem is they can’t pay for it, even if it’s their fault for not having insurance. Let me give a parallel. A number of years ago in the Sierra Club bulletin there was a letter that had to do with the fact that the National Park Service was proposing a rule that every climber had to have a helmet. This fellow, who had been a famous mountain climber and was then a manufacturer of very-high-quality climbing equipment, said he objected to it on grounds of freedom. He said he understood the reason for it, that if a person is trapped or injured but has his head intact, he can cooperate with the rescuers and make the rescue much easier if he has a helmet. He said, therefore, the only proposition is that if you go climbing without a helmet, you understand that you are not going to be rescued. Now it shocked me. There is somehow a moral judgment here, that we can just let people die; therefore the financing comes through the tax system, because of unequal distribution. You could think of, although it’s difficult, an adequate insurance system where the ability to buy insurance would be related to income. Let’s for the moment imagine we can overcome all these moral hazard issues and we can have really good Q

78

Q

DISCUSSION

insurance serving the proper function. Nevertheless, we would have the problem that after all, health costs are by no means proportional to income. If you take the average health cost at about $8,000 per person, it is obvious that an insurance adequate to cover that would not be sustainable, so we have Medicaid. In other words, if we take an economic point of view, we can say that what we have is a lot of taxation, and taxation is distortionary. The costs are really higher than the money costs because there is a classic welfare loss due to excess burden due to taxation. It seems to me that is why all these special problems arise, because we saying we’re going to pay for it by taxes some way or another. Taxes mean a bigger role for the government, which we know is somewhat unpopular in some quarters, but even from a purely economic point of view, it does involve welfare losses, which have to be offset against what we think are the gains. Interestingly, nobody is going to repeal a lot of this stuff. The marginal group of uninsured is only 50 million, that’s why we can talk about throwing them to the dogs. We’re not going to destroy their Medicare, and in fact there is no great opposition to Medicaid. Maybe there is opposition on the margin, but it’s surprising to me that given the diversity of opinion, this is accepted. That is why it becomes a big issue, I think; it might not have if it was just a question of increased market share. STIGLITZ: Just two comments: one, we’ve talked about the incentives of the providers, and one of the interesting aspects of that is the peculiarity of fee-for-provider, where you have an incentive to sell more. The other one of course Q

79

Q

DISCUSSION

is the HMOs, where they get per capitation, and they have an incentive to provide nothing. We hope that the reputation mechanism works, but we know that reputation mechanisms don’t work very well, particularly when choices are limited. I mentioned that because it highlights the difficulty of solving the incentive problem; if it were easy we would have solved it by now. It is clear that other countries have done a better job of this, and part of it may be a sense of professional responsibilities, the kind of ethical things that Ken was talking about. The second thing is that we have been focusing on the health care sector, but we have said actually relatively little about health. We ought always to remember that the health care sector is an input to something that we really care about, which is health, and it’s not the most important input. Other things like nutrition, smoking, and drinking probably have more important effects on average on a lot of aspects of our health than does health care. The reason why this is partly an economic problem, of course, is that if we are going to subsidize the provision of medical care, it also raises the question of our positions about taxing and subsidizing activities that affect health.

Q

80

Q

ARROW (1963): UNCERTAINTY AND THE WELFARE ECONOMICS OF MEDICAL CARE BY KENNETH J. ARROW*

I. INTRODUCTION: SCOPE AND METHOD This paper is an exploratory and tentative study of the specific differentia of medical care as the object of normative economics. It is contended here, on the basis of comparison of obvious characteristics of the medical-care industry with the norms of welfare economics, that the special economic problems of medical care can be explained as adaptations to the existence of uncertainty in the incidence of disease and in the efficacy of treatment. It should be noted that the subject is the medical-care industry, not health. The causal factors in health are many, and the provision of medical care is only one. Particularly at low levels of income, other commodities such as nutrition, shelter, clothing, and sanitation may be much more significant. It is the complex of services that center about the

Reprinted by permission from the American Economic Review, 53(5), 941–973. Copyright © 1963 American Economic Association.

ARROW (1963)

physician, private and group practice, hospitals, and public health, which I propose to discuss. The focus of discussion will be on the way the operation of the medical-care industry and the efficacy with which it satisfies the needs of society differ from a norm, if at all. The “norm” that the economist usually uses for the purposes of such comparisons is the operation of a competitive model, that is, the flows of services that would be offered and purchased and the prices that would be paid for them if each individual in the market offered or purchased services at the going prices as if his decisions had no influence over them, and the going prices were such that the amounts of services which were available equalled the total amounts which other individuals were willing to purchase, with no imposed restrictions on supply or demand. The interest in the competitive model stems partly from its presumed descriptive power and partly from its implications for economic efficiency. In particular, we can state the following well-known proposition (First Optimality Theorem). If a competitive equilibrium exists at all, and if all commodities relevant to costs or utilities are in fact priced in the market, then the equilibrium is necessarily optimal in the following precise sense (due to V. Pareto): There is no other allocation of resources to services which will make all participants in the market better off. Both the conditions of this optimality theorem and the definition of optimality call for comment. A definition is just a definition, but when the definiendum is a word already in common use with highly favorable connotations, it is clear that we are really trying to be persuasive; we are implicitly Q

82

Q

ARROW (1963)

recommending the achievement of optimal states.1 It is reasonable enough to assert that a change in allocation which makes all participants better off is one that certainly should be made; this is a value judgment, not a descriptive proposition, but it is a very weak one. From this it follows that it is not desirable to put up with a non-optimal allocation. But it does not follow that if we are at an allocation which is optimal in the Pareto sense, we should not change to any other. We cannot indeed make a change that does not hurt someone; but we can still desire to change to another allocation if the change makes enough participants better off and by so much that we feel that the injury to others is not enough to offset the benefits. Such interpersonal comparisons are, of course, value judgments. The change, however, by the previous argument ought to be an optimal state; of course there are many possible states, each of which is optimal in the sense here used. However, a value judgment on the desirability of each possible new distribution of benefits and costs corresponding to each possible reallocation of resources is not, in general, necessary. Judgments about the distribution can be made separately, in one sense, from those about allocation if certain conditions are fulfilled. Before stating the relevant proposition, it is necessary to remark that the competitive equilibrium achieved depends in good measure on the initial distribution of purchasing power, which consists of ownership of assets and skills that command a price on the market. A transfer of assets among individuals will, in general, change the final supplies of goods and services and the prices paid for them. Thus, a transfer of purchasing power from the Q

83

Q

ARROW (1963)

well to the ill will increase the demand for medical services. This will manifest itself in the short run in an increase in the price of medical services and in the long run in an increase in the amount supplied. With this in mind, the following statement can be made (Second Optimality Theorem): If there are no increasing returns in production, and if certain other minor conditions are satisfied, then every optimal state is a competitive equilibrium corresponding to some initial distribution of purchasing power. Operationally, the significance of this proposition is that if the conditions of the two optimality theorems are satisfied, and if the allocation mechanism in the real world satisfies the conditions for a competitive model, then social policy can confine itself to steps taken to alter the distribution of purchasing power. For any given distribution of purchasing power, the market will, under the assumptions made, achieve a competitive equilibrium which is necessarily optimal; and any optimal state is a competitive equilibrium corresponding to some distribution of purchasing power, so that any desired optimal state can be achieved. The redistribution of purchasing power among individuals most simply takes the form of money: taxes and subsidies. The implications of such a transfer for individual satisfactions are, in general, not known in advance. But we can assume that society can ex post judge the distribution of satisfactions and, if deemed unsatisfactory, take steps to correct it by subsequent transfers. Thus, by successive approximations, a most preferred social state can be achieved, with resource allocation being handled by the market and public policy confined to the redistribution of money income.2 Q

84

Q

ARROW (1963)

If, on the contrary, the actual market differs significantly from the competitive model, or if the assumptions of the two optimality theorems are not fulfilled, the separation of allocative and distributional procedures becomes, in most cases, impossible.3 The first step then in the analysis of the medical-care market is the comparison between the actual market and the competitive model. The methodology of this comparison has been a recurrent subject of controversy in economics for over a century. Recently, M. Friedman [15] has vigorously argued that the competitive or any other model should be tested solely by its ability to predict. In the context of competition, he comes close to arguing that prices and quantities are the only relevant data. This point of view is valuable in stressing that a certain amount of lack of realism in the assumptions of a model is no argument against its value. But the price-quantity implications of the competitive model for pricing are not easy to derive without major—and, in many cases, impossible—econometric efforts. In this paper, the institutional organization and the observable mores of the medical profession are included among the data to be used in assessing the competitiveness of the medical-care market. I shall also examine the presence or absence of the preconditions for the equivalence of competitive equilibria and optimal states. The major competitive preconditions, in the sense used here, are three: the existence of competitive equilibrium, the marketability of all goods and services relevant to costs and utilities, and nonincreasing returns. The first two, as we have seen, insure that competitive equilibrium is necessarily optimal; the third insures Q

85

Q

ARROW (1963)

that every optimal state is the competitive equilibrium corresponding to some distribution of income.4 The first and third conditions are interrelated; indeed, nonincreasing returns plus some additional conditions not restrictive in a modern economy imply the existence of a competitive equilibrium, i.e., imply that there will be some set of prices which will clear all markets.5 The concept of marketability is somewhat broader than the traditional divergence between private and social costs and benefits. The latter concept refers to cases in which the organization of the market does not require an individual to pay for costs that he imposes on others as the result of his actions or does not permit him to receive compensation for benefits he confers. In the medical field, the obvious example is the spread of communicable diseases. An individual who fails to be immunized not only risks his own health, a disutility which presumably he has weighed against the utility of avoiding the procedure, but also that of others. In an ideal price system, there would be a price which he would have to pay to anyone whose health is endangered, a price sufficiently high so that the others would feel compensated; or, alternatively, there would be a price which would be paid to him by others to induce him to undergo the immunization procedure. Either system would lead to an optimal state, though the distributional implications would be different. It is, of course, not hard to see that such price systems could not, in fact, be practical; to approximate an optimal state it would be necessary to have collective intervention in the form of subsidy or tax or compulsion. Q

86

Q

ARROW (1963)

By the absence of marketability for an action which is identifiable, technologically possible, and capable of influencing some individual’s welfare, for better or for worse, is meant here the failure of the existing market to provide a means whereby the services can be both offered and demanded upon payment of a price. Nonmarketability may be due to intrinsic technological characteristics of the product which prevent a suitable price from being enforced, as in the case of communicable diseases, or it may be due to social or historical controls, such as those prohibiting an individual from selling himself into slavery. This distinction is, in fact, difficult to make precise, though it is obviously of importance for policy; for the present purposes, it will be sufficient to identify nonmarketability with the observed absence of markets. The instance of nonmarketability with which we shall be most concerned is that of risk-bearing. The relevance of riskbearing to medical care seems obvious; illness is to a considerable extent an unpredictable phenomenon. The ability to shift the risks of illness to others is worth a price which many are willing to pay. Because of pooling and of superior willingness and ability, others are willing to bear the risks. Nevertheless, as we shall see in greater detail, a great many risks are not covered, and indeed the markets for the services of risk-coverage are poorly developed or nonexistent. Why this should be so is explained in more detail in Section IV.C below; briefly, it is impossible to draw up insurance policies which will sufficiently distinguish among risks, particularly since observation of the results will be incapable of distinguishing between avoidable and unavoidable risks, so that incentives to avoid losses are diluted. Q

87

Q

ARROW (1963)

The optimality theorems discussed above are usually presented in the literature as referring only to conditions of certainty, but there is no difficulty in extending them to the case of risks, provided the additional services of risk-bearing are included with other commodities.6 However, the variety of possible risks in the world is really staggering. The relevant commodities include, in effect, bets on all possible occurrences in the world which impinge upon utilities. In fact, many of these “commodities,” i.e., desired protection against many risks, are simply not available. Thus, a wide class of commodities is nonmarket- able, and a basic competitive precondition is not satisfied.7 There is a still more subtle consequence of the introduction of risk-bearing considerations. When there is uncertainty, information or knowledge becomes a commodity. Like other commodities, it has a cost of production and a cost of transmission, and so it is naturally not spread out over the entire population but concentrated among those who can profit most from it. (These costs may be measured in time or disutility as well as money.) But the demand for information is difficult to discuss in the rational terms usually employed. The value of information is frequently not known in any meaningful sense to the buyer; if, indeed, he knew enough to measure the value of information, he would know the information itself. But information, in the form of skilled care, is precisely what is being bought from most physicians, and, indeed, from most professionals. The elusive character of information as a commodity suggests that it departs considerably from the usual marketability assumptions about commodities.8 Q

88

Q

ARROW (1963)

That risk and uncertainty are, in fact, significant elements in medical care hardly needs argument. I will hold that virtually all the special features of this industry, in fact, stem from the prevalence of uncertainty. The nonexistence of markets for the bearing of some risks in the first instance reduces welfare for those who wish to transfer those risks to others for a certain price, as well as for those who would find it profitable to take on the risk at such prices. But it also reduces the desire to render or consume services which have risky consequences; in technical language, these commodities are complementary to risk-bearing. Conversely, the production and consumption of commodities and services with little risk attached act as substitutes for risk-bearing and are encouraged by market failure there with respect to risk-bearing. Thus the observed commodity pattern will be affected by the nonexistence of other markets. The failure of one or more of the competitive preconditions has as its most immediate and obvious consequence a reduction in welfare below that obtainable from existing resources and technology, in the sense of a failure to reach an optimal state in the sense of Pareto. But more can be said. I propose here the view that, when the market fails to achieve an optimal state, society will, to some extent at least, recognize the gap, and nonmarket social institutions will arise attempting to bridge it.9 Certainly this process is not necessarily conscious; nor is it uniformly successful in approaching more closely to optimality when the entire range of consequences is considered. It has always been a favorite activity of economists to point out that actions which on their face achieve a desirable goal may have less Q

89

Q

ARROW (1963)

obvious consequences, particularly over time, which more than offset the original gains. But it is contended here that the special structural characteristics of the medical-care market are largely attempts to overcome the lack of optimality due to the nonmarketability of the bearing of suitable risks and the imperfect marketability of information. These compensatory institutional changes, with some reinforcement from usual profit motives, largely explain the observed noncompetitive behavior of the medical-care market, behavior which, in itself, interferes with optimality. The social adjustment towards optimality thus puts obstacles in its own path. The doctrine that society will seek to achieve optimality by non-market means if it cannot achieve them in the market is not novel. Certainly, the government, at least in its economic activities, is usually implicitly or explicitly held to function as the agency which substitutes for the market’s failure.10 I am arguing here that in some circumstances other social institutions will step into the optimality gap, and that the medical-care industry, with its variety of special institutions, some ancient, some modern, exemplifies this tendency. It may be useful to remark here that a good part of the preference for redistribution expressed in government taxation and expenditure policies and private charity can be reinterpreted as desire for insurance. It is noteworthy that virtually nowhere is there a system of subsidies that has as its aim simply an equalization of income. The subsidies or other governmental help go to those who are disadvantaged in life by events the incidence of which is popularly regarded as unpredictable: the blind, dependent children, Q

90

Q

ARROW (1963)

the medically indigent. Thus, optimality, in a context which includes risk-bearing, includes much that appears to be motivated by distributional value judgments when looked at in a narrower context.11 This methodological background gives rise to the following plan for this paper. Section II is a catalogue of stylized generalizations about the medical-care market which differentiate it from the usual commodity markets. In Section III the behavior of the market is compared with that of the competitive model which disregards the fact of uncertainty. In Section IV, the medical-care market is compared, both as to behavior and as to preconditions, with the ideal competitive market that takes account of uncertainty; an attempt will be made to demonstrate that the characteristics outlined in Section II can be explained either as the result of deviations from the competitive preconditions or as attempts to compensate by other institutions for these failures. The discussion is not designed to be definitive, but provocative. In particular, I have been chary about drawing policy inferences; to a considerable extent, they depend on further research, for which the present paper is intended to provide a framework.

II. A SURVEY OF THE SPECIAL CHARACTERISTICS OF THE MEDICAL-CARE MARKET12 This section will list selectively some characteristics of medical care which distinguish it from the usual commodity of economics textbooks. The list is not exhaustive, and it is Q

91

Q

ARROW (1963)

not claimed that the characteristics listed are individually unique to this market. But, taken together, they do establish a special place for medical care in economic analysis.

A. THE NATURE OF DEMAND The most obvious distinguishing characteristics of an individual’s demand for medical services is that it is not steady in origin as, for example, for food or clothing, but irregular and unpredictable. Medical services, apart from preventive services, afford satisfaction only in the event of illness, a departure from the normal state of affairs. It is hard, indeed, to think of another commodity of significance in the average budget of which this is true. A portion of legal services, devoted to defense in criminal trials or to lawsuits, might fall in this category but the incidence is surely very much lower (and, of course, there are, in fact, strong institutional similarities between the legal and medical-care markets.)13 In addition, the demand for medical services is associated, with a considerable probability, with an assault on personal integrity. There is some risk of death and a more considerable risk of impairment of full functioning. In particular, there is a major potential for loss or reduction of earning ability. The risks are not by themselves unique; food is also a necessity, but avoidance of deprivation of food can be guaranteed with sufficient income, where the same cannot be said of avoidance of illness. Illness is, thus, not only risky but a costly risk in itself, apart from the cost of medical care. Q

92

Q

ARROW (1963)

B. EXPECTED BEHAVIOR OF THE PHYSICIAN It is clear from everyday observation that the behavior expected of sellers of medical care is different from that of business men in general. These expectations are relevant because medical care belongs to the category of commodities for which the product and the activity of production are identical. In all such cases, the customer cannot test the product before consuming it, and there is an element of trust in the relation.14 But the ethically understood restrictions on the activities of a physician are much more severe than on those of, say, a barber. His behavior is supposed to be governed by a concern for the customer’s welfare which would not be expected of a salesman. In Talcott Parsons’s terms, there is a “collectivity-orientation,” which distinguishes medicine and other professions from business, where selfinterest on the part of participants is the accepted norm.15 A few illustrations will indicate the degree of difference between the behavior expected of physicians and that expected of the typical businessman.16 (1) Advertising and overt price competition are virtually eliminated among physicians. (2) Advice given by physicians as to further treatment by himself or others is supposed to be completely divorced from self-interest. (3) It is at least claimed that treatment is dictated by the objective needs of the case and not limited by financial considerations.17 While the ethical compulsion is surely not as absolute in fact as it is in theory, we can hardly suppose that it has no influence over resource allocation in this area. Charity treatment in one form or another does exist because of this tradition about Q

93

Q

ARROW (1963)

human rights to adequate medical care.18 (4) The physician is relied on as an expert in certifying to the existence of illnesses and injuries for various legal and other purposes. It is socially expected that his concern for the correct conveying of information will, when appropriate, outweigh his desire to please his customers.19 Departure from the profit motive is strikingly manifested by the overwhelming predominance of nonprofit over proprietary hospitals.20 The hospital per se offers services not too different from those of a hotel, and it is certainly not obvious that the profit motive will not lead to a more efficient supply. The explanation may lie either on the supply side or on that of demand. The simplest explanation is that public and private subsidies decrease the cost to the patient in nonprofit hospitals. A second possibility is that the association of profit-making with the supply of medical services arouses suspicion and antagonism on the part of patients and referring physicians, so they do prefer nonprofit institutions. Either explanation implies a preference on the part of some group, whether donors or patients, against the profit motive in the supply of hospital services.21 Conformity to collectivity-oriented behavior is especially important since it is a commonplace that the physicianpatient relation affects the quality of the medical care product. A pure cash nexus would be inadequate; if nothing else, the patient expects that the same physician will normally treat him on successive occasions. This expectation is strong enough to persist even in the Soviet Union, where medical care is nominally removed from the market place [14, pp. 194– 96]. That purely psychic interactions between physician and Q

94

Q

ARROW (1963)

patient have effects which are objectively indistinguishable in kind from the effects of medication is evidenced by the use of the placebo as a control in medical experimentation; see Shapiro [25].

C. PRODUCT UNCERTAINTY Uncertainty as to the quality of the product is perhaps more intense here than in any other important commodity. Recovery from disease is as unpredictable as is its incidence. In most commodities, the possibility of learning from one’s own experience or that of others is strong because there is an adequate number of trials. In the case of severe illness, that is, in general, not true; the uncertainty due to inexperience is added to the intrinsic difficulty of prediction. Further, the amount of uncertainty, measured in terms of utility variability, is certainly much greater for medical care in severe cases than for, say, houses or automobiles, even though these are also expenditures sufficiently infrequent so that there may be considerable residual uncertainty. Further, there is a special quality to the uncertainty; it is very different on the two sides of the transaction. Because medical knowledge is so complicated, the information possessed by the physician as to the consequences and possibilities of treatment is necessarily very much greater than that of the patient, or at least so it is believed by both parties.22 Further, both parties are aware of this informational inequality, and their relation is colored by this knowledge. To avoid misunderstanding, observe that the difference in information relevant here is a difference in information as Q

95

Q

ARROW (1963)

to the consequence of a purchase of medical care. There is always an inequality of information as to production methods between the producer and the purchaser of any commodity, but in most cases the customer may well have as good or nearly as good an understanding of the utility of the product as the producer.

D. SUPPLY CONDITIONS In competitive theory, the supply of a commodity is governed by the net return from its production compared with the return derivable from the use of the same resources elsewhere. There are several significant departures from this theory in the case of medical care. Most obviously, entry to the profession is restricted by licensing. Licensing, of course, restricts supply and therefore increases the cost of medical care. It is defended as guaranteeing a minimum of quality. Restriction of entry by licensing occurs in most professions, including barbering and undertaking. A second feature is perhaps even more remarkable. The cost of medical education today is high and, according to the usual figures, is borne only to a minor extent by the student. Thus, the private benefits to the entering student considerably exceed the costs. (It is, however, possible that research costs, not properly chargeable to education, swell the apparent difference.) This subsidy should, in principle, cause a fall in the price of medical services, which, however, is offset by rationing through limited entry to schools and through elimination of students during the medical-school career. Q

96

Q

ARROW (1963)

These restrictions basically render superfluous the licensing, except in regard to graduates of foreign schools. The special role of educational institutions in simultaneously subsidizing and rationing entry is common to all professions requiring advanced training.23 It is a striking and insufficiently remarked phenomenon that such an important part of resource allocation should be performed by nonprofit-oriented agencies. Since this last phenomenon goes well beyond the purely medical aspect, we will not dwell on it longer here except to note that the anomaly is most striking in the medical field. Educational costs tend to be far higher there than in any other branch of professional training. While tuition is the same, or only slightly higher, so that the subsidy is much greater, at the same time the earnings of physicians rank highest among professional groups, so there would not at first blush seem to be any necessity for special inducements to enter the profession. Even if we grant that, for reasons unexamined here, there is a social interest in subsidized professional education, it is not clear why the rate of subsidization should differ among professions. One might expect that the tuition of medical students would be higher than that of other students. The high cost of medical education in the United States is itself a reflection of the quality standards imposed by the American Medical Association since the Flexner Report, and it is, I believe, only since then that the subsidy element in medical education has become significant. Previously, many medical schools paid their way or even yielded a profit. Another interesting feature of limitation on entry to subsidized education is the extent of individual preferences Q

97

Q

ARROW (1963)

concerning the social welfare, as manifested by contributions to private universities. But whether support is public or private, the important point is that both the quality and the quantity of the supply of medical care are being strongly influenced by social nonmarket forces.24,25 One striking consequence of the control of quality is the restriction on the range offered. If many qualities of a commodity are possible, it would usually happen in a competitive market that many qualities will be offered on the market, at suitably varying prices, to appeal to different tastes and incomes. Both the licensing laws and the standards of medicalschool training have limited the possibilities of alternative qualities of medical care. The declining ratio of physicians to total employees in the medical-care industry shows that substitution of less trained personnel, technicians, and the like, is not prevented completely, but the central role of the highly trained physician is not affected at all.26

E. PRICING PRACTICES The unusual pricing practices and attitudes of the medical profession are well known: extensive price discrimination by income (with an extreme of zero prices for sufficiently indigent patients) and, formerly, a strong insistence on fee for services as against such alternatives as prepayment. The opposition to prepayment is closely related to an even stronger opposition to closed-panel practice (contractual arrangements which bind the patient to a particular group of physicians). Again these attitudes seem to differentiate professions from business. Prepayment and closed-panel Q

98

Q

ARROW (1963)

plans are virtually nonexistent in the legal profession. In ordinary business, on the other hand, there exists a wide variety of exclusive service contracts involving sharing of risks; it is assumed that competition will select those which satisfy needs best.27 The problems of implicit and explicit price-fixing should also be mentioned. Price competition is frowned on. Arrangements of this type are not uncommon in service industries, and they have not been subjected to antitrust action. How important this is is hard to assess. It has been pointed out many times that the apparent rigidity of so-called administered prices considerably understates the actual flexibility. Here, too, if physicians find themselves with unoccupied time, rates are likely to go down, openly or covertly; if there is insufficient time for the demand, rates will surely rise. The “ethics” of price competition may decrease the flexibility of price responses, but probably that is all.

III. COMPARISONS WITH THE COMPETITIVE MODEL UNDER CERTAINTY A. NONMARKETABLE COMMODITIES As already noted, the diffusion of communicable diseases provides an obvious example of nonmarket interactions. But from a theoretical viewpoint, the issues are well understood, and there is little point in expanding on this theme. (This should not be interpreted as minimizing the contribution of public health to welfare; there is every reason to suppose Q

99

Q

ARROW (1963)

that it is considerably more important than all other aspects of medical care.) Beyond this special area there is a more general interdependence, the concern of individuals for the health of others. The economic manifestations of this taste are to be found in individual donations to hospitals and to medical education, as well as in the widely accepted responsibilities of government in this area. The taste for improving the health of others appears to be stronger than for improving other aspects of their welfare.28 In interdependencies generated by concern for the welfare of others there is always a theoretical case for collective action if each participant derives satisfaction from the contributions of all.

B. INCREASING RETURNS Problems associated with increasing returns play some role in allocation of resources in the medical field, particularly in areas of low density or low income. Hospitals show increasing returns up to a point; specialists and some medical equipment constitute significant indivisibilities. In many parts of the world the individual physician may be a large unit relative to demand. In such cases it can be socially desirable to subsidize the appropriate medical-care unit. The appropriate mode of analysis is much the same as for water-resource projects. Increasing returns are hardly apt to be a significant problem in general practice in large cities in the United States, and improved transportation to some extent reduces their importance elsewhere. Q

100

Q

ARROW (1963)

C. ENTRY The most striking departure from competitive behavior is restriction on entry to the field, as discussed in II.D above. Friedman and Kuznets, in a detailed examination of the preWorld War II data, have argued that the higher income of physicians could be attributed to this restriction.29 There is some evidence that the demand for admission to medical school has dropped (as indicated by the number of applicants per place and the quality of those admitted), so that the number of medical-school places is not as significant a barrier to entry as in the early 1950’s [28, pp. 14–15], But it certainly has operated over the past and it is still operating to a considerable extent today. It has, of course, constituted a direct and unsubtle restriction on the supply of medical care. There are several considerations that must be added to help evaluate the importance of entry restrictions: (1) Additional entrants would be, in general, of lower quality; hence, the addition to the supply of medical care, properly adjusted for quality, is less than purely quantitative calculations would show.30 (2) To achieve genuinely competitive conditions, it would be necessary not only to remove numerical restrictions on entry but also to remove the subsidy in medical education. Like any other producer, the physician should bear all the costs of production, including, in this case, education.31 It is not so clear that this change would not keep even unrestricted entry down below the present level. (3) To some extent, the effect of making tuition carry the full cost of education will be to create too few entrants, rather than too many. Given the imperfections of the capital market, loans Q

101

Q

ARROW (1963)

for this purpose to those who do not have the cash are difficult to obtain. The lender really has no security. The obvious answer is some form of insured loans, as has frequently been argued; not too much ingenuity would be needed to create a credit system for medical (and other branches of higher) education. Under these conditions the cost would still constitute a deterrent, but one to be compared with the high future incomes to be obtained. If entry were governed by ideal competitive conditions, it may be that the quantity on balance would be increased, though this conclusion is not obvious. The average quality would probably fall, even under an ideal credit system, since subsidy plus selected entry draw some highly qualified individuals who would otherwise get into other fields. The decline in quality is not an over-all social loss, since it is accompanied by increase in quality in other fields of endeavor; indeed, if demands accurately reflected utilities, there would be a net social gain through a switch to competitive entry.32 There is a second aspect of entry in which the contrast with competitive behavior is, in many respects, even sharper. It is the exclusion of many imperfect substitutes for physicians. The licensing laws, though they do not effectively limit the number of physicians, do exclude all others from engaging in any one of the activities known as medical practice. As a result, costly physician time may be employed at specific tasks for which only a small fraction of their training is needed, and which could be performed by others less well trained and therefore less expensive. One might expect immunization centers, privately operated, but not necessarily requiring the services of doctors. Q

102

Q

ARROW (1963)

In the competitive model without uncertainty, consumers are presumed to be able to distinguish qualities of the commodities they buy. Under this hypothesis, licensing would be, at best, superfluous and exclude those from whom consumers would not buy anyway; but it might exclude too many.

D. PRICING The pricing practices of the medical industry (see II.E above) depart sharply from the competitive norm. As Kessel [17] has pointed out with great vigor, not only is price discrimination incompatible with the competitive model, but its preservation in the face of the large number of physicians is equivalent to a collective monopoly. In the past, the opposition to prepayment plans has taken distinctly coercive forms, certainly transcending market pressures, to say the least. Kessel has argued that price discrimination is designed to maximize profits along the classic lines of discriminating monopoly and that organized medical opposition to prepayment was motivated by the desire to protect these profits. In principle, prepayment schemes are compatible with discrimination, but in practice they do not usually discriminate. I do not believe the evidence that the actual scale of discrimination is profit-maximizing is convincing. In particular, note that for any monopoly, discriminating or otherwise, the elasticity of demand in each market at the point of maximum profits is greater than one. But it is almost surely true for medical care that the price elasticity of demand for all income levels is less than one. That price discrimination by income is not completely profit-maximizing is obvious in Q

103

Q

ARROW (1963)

the extreme case of charity; Kessel argues that this represents an appeasement of public opinion. But this already shows the incompleteness of the model and suggests the relevance and importance of social and ethical factors. Certainly one important part of the opposition to prepayment was its close relation to closed-panel plans. Prepayment is a form of insurance, and naturally the individual physician did not wish to assume the risks. Pooling was intrinsically involved, and this strongly motivates, as we shall discuss further in Section IV below, control over prices and benefits. The simplest administrative form is the closed panel; physicians involved are, in effect, the insuring agent. From this point of view, Blue Cross solved the prepayment problem by universalizing the closed panel. The case that price discrimination by income is a form of profit maximization which was zealously defended by opposition to fees for service seems far from proven. But it remains true that this price discrimination, for whatever cause, is a source of nonoptimality. Hypothetically, it means everyone would be better off if prices were made equal for all, and the rich compensated the poor for the changes in the relative positions. The importance of this welfare loss depends on the actual amount of discrimination and on the elasticities of demand for medical services by the different income groups. If the discussion is simplified by considering only two income levels, rich and poor, and if the elasticity of demand by either one is zero, then no reallocation of medical services will take place and the initial situation is optimal. The only effect of a change in price will be the redistribution of income as between the medical profession and the Q

104

Q

ARROW (1963)

group with the zero elasticity of demand. With low elasticities of demand, the gain will be small. To illustrate, suppose the price of medical care to the rich is double that to the poor, the medical expenditures by the rich are 20 per cent of those by the poor, and the elasticity of demand for both classes is .5; then the net social gain due to the abolition of discrimination is slightly over 1 per cent of previous medical expenditures.33 The issues involved in the opposition to prepayment, the other major anomaly in medical pricing, are not meaningful in the world of certainty and will be discussed below.

IV. COMPARISON WITH THE IDEAL COMPETITIVE MODEL UNDER UNCERTAINTY A. INTRODUCTION In this section we will compare the operations of the actual medical-care market with those of an ideal system in which not only the usual commodities and services but also insurance policies against all conceivable risks are available.34 Departures consist for the most part of insurance policies that might conceivably be written, but are in fact not. Whether these potential commodities are nonmarketable, or, merely because of some imperfection in the market, are not actually marketed, is a somewhat fine point. To recall what has already been said in Section I, there are two kinds of risks involved in medical care: the risk of becoming ill, and the risk of total or incomplete or delayed Q

105

Q

ARROW (1963)

recovery. The loss due to illness is only partially the cost of medical care. It also consists of discomfort and loss of productive time during illness, and, in more serious cases, death or prolonged deprivation of normal function. From the point of view of the welfare economics of uncertainty, both losses are risks against which individuals would like to insure. The nonexistence of suitable insurance policies for either risk implies a loss of welfare.

B. THE THEORY OF IDEAL INSURANCE In this section, the basic principles of an optimal regime for risk-bearing will be presented. For illustration, reference will usually be made to the case of insurance against cost in medical care. The principles are equally applicable to any of the risks. There is no single source to which the reader can be easily referred, though I think the principles are at least reasonably well understood. As a basis for the analysis, the assumption is made that each individual acts so as to maximize the expected value of a utility function. If we think of utility as attached to income, then the costs of medical care act as a random deduction from this income, and it is the expected value of the utility of income after medical costs that we are concerned with. (Income after medical costs is the ability to spend money on other objects which give satisfaction. We presuppose that illness is not a source of satisfaction in itself; to the extent that it is a source of dissatisfaction, the illness should enter into the utility function as a separate variable.) The expectedutility hypothesis, due originally to Daniel Bernoulli (1738), Q

106

Q

ARROW (1963)

is plausible and is the most analytically manageable of all hypotheses that have been proposed to explain behavior under uncertainty. In any case, the results to follow probably would not be significantly affected by moving to another mode of analysis. It is further assumed that individuals are normally riskaverters. In utility terms, this means that they have a diminishing marginal utility of income. This assumption may reasonably be taken to hold for most of the significant affairs of life for a majority of people, but the presence of gambling provides some difficulty in the full application of this view. It follows from the assumption of risk aversion that if an individual is given a choice between a probability distribution of income, with a given mean m, and the certainty of the income m, he would prefer the latter. Suppose, therefore, an agency, a large insurance company plan, or the government, stands ready to offer insurance against medical costs on an actuarially fair basis; that is, if the costs of medical care are a random variable with mean m, the company will charge a premium m, and agree to indemnify the individual for all medical costs. Under these circumstances, the individual will certainly prefer to take out a policy and will have a welfare gain thereby. Will this be a social gain? Obviously yes, if the insurance agent is suffering no social loss. Under the assumption that medical risks on different individuals are basically independent, the pooling of them reduces the risk involved to the insurer to relatively small proportions. In the limit, the welfare loss, even assuming risk aversion on the part of the insurer, would vanish and there is a net social gain which may be of quite substantial magnitude. In fact, of course, the Q

107

Q

ARROW (1963)

pooling of risks does not go to the limit; there is only a finite number of them and there may be some interdependence among the risks due to epidemics and the like. But then a premium, perhaps slightly above the actuarial level, would be sufficient to offset this welfare loss. From the point of view of the individual, since he has a strict preference for the actuarially fair policy over assuming the risks himself, he will still have a preference for an actuarially unfair policy, provided, of course, that it is not too unfair. In addition to a residual degree of risk aversion by insurers, there are other reasons for the loading of the premium (i.e., an excess of premium over the actuarial value). Insurance involves administrative costs. Also, because of the irregularity of payments there is likely to be a cost of capital tied up. Suppose, to take a simple case, the insurance company is not willing to sell any insurance policy that a consumer wants but will charge a fixed-percentage loading above the actuarial value for its premium. Then it can be shown that the most preferred policy from the point of view of an individual is a coverage with a deductible amount; that is, the insurance policy provides 100 per cent coverage for all medical costs in excess of some fixed-dollar limit. If, however, the insurance company has some degree of risk aversion, its loading may also depend on the degree of uncertainty of the risk. In that case, the Pareto optimal policy will involve some element of co-insurance, i.e., the coverage for costs over the minimum limit will be some fraction less than 100 per cent (for proofs of these statements, see Appendix). These results can also be applied to the hypothetical concept of insurance against failure to recover from illness. For Q

108

Q

ARROW (1963)

simplicity, let us assume that the cost of failure to recover is regarded purely as a money cost, either simply productive opportunities foregone or, more generally, the money equivalent of all dissatisfactions. Suppose further that, given that a person is ill, the expected value of medical care is greater than its cost; that is, the expected money value attributable to recovery with medical help is greater than resources devoted to medical help. However, the recovery, though on the average beneficial, is uncertain; in the absence of insurance a risk-averter may well prefer not to take a chance on further impoverishment by buying medical care. A suitable insurance policy would, however, mean that he paid nothing if he doesn’t benefit; since the expected value is greater than the cost, there would be a net social gain.35

C. PROBLEMS OF INSURANCE 1. The moral hazard. The welfare case for insurance policies of all sorts is overwhelming. It follows that the government should undertake insurance in those cases where this market, for whatever reason, has failed to emerge. Nevertheless, there are a number of significant practical limitations on the use of insurance. It is important to understand them, though I do not believe that they alter the case for the creation of a much wider class of insurance policies than now exists. One of the limits which has been much stressed in insurance literature is the effect of insurance on incentives. What is desired in the case of insurance is that the event against which insurance is taken be out of the control of the individual. Unfortunately, in real life this separation can never Q

109

Q

ARROW (1963)

be made perfectly. The outbreak of fire in one’s house or business may be largely uncontrollable by the individual, but the probability of fire is somewhat influenced by carelessness, and of course arson is a possibility, if an extreme one. Similarly, in medical policies the cost of medical care is not completely determined by the illness suffered by the individual but depends on the choice of a doctor and his willingness to use medical services. It is frequently observed that widespread medical insurance increases the demand for medical care. Coinsurance provisions have been introduced into many major medical policies to meet this contingency as well as the risk aversion of the insurance companies. To some extent the professional relationship between physician and patient limits the normal hazard in various forms of medical insurance. By certifying to the necessity of given treatment or the lack thereof, the physician acts as a controlling agent on behalf of the insurance companies. Needless to say, it is a far from perfect check; the physicians themselves are not under any control and it may be convenient for them or pleasing to their patients to prescribe more expensive medication, private nurses, more frequent treatments, and other marginal variations of care. It is probably true that hospitalization and surgery are more under the casual inspection of others than is general practice and therefore less subject to moral hazard; this may be one reason why insurance policies in those fields have been more widespread. 2. Alternative methods of insurance payment. It is interesting that no less than three different methods of coverage of the costs of medical care have arisen: prepayment, indemnities according to a fixed schedule, and insurance against Q

110

Q

ARROW (1963)

costs, whatever they may be. In prepayment plans, insurance in effect is paid in kind—that is, directly in medical services. The other two forms both involve cash payments to the beneficiary, but in the one case the amounts to be paid involving a medical contingency are fixed in advance, while in the other the insurance carrier pays all the costs, whatever they may be, subject, of course, to provisions like deductibles and coinsurance. In hypothetically perfect markets these three forms of insurance would be equivalent. The indemnities stipulated would, in fact, equal the market price of the services, so that value to the insured would be the same if he were to be paid the fixed sum or the market price or were given the services free. In fact, of course, insurance against full costs and prepayment plans both offer insurance against uncertainty as to the price of medical services, in addition to uncertainty about their needs. Further, by their mode of compensation to the physician, prepayment plans are inevitably bound up with closed panels so that the freedom of choice of the physician by the patient is less than it would be under a scheme more strictly confined to the provision of insurance. These remarks are tentative, and the question of coexistence of the different schemes should be a fruitful subject for investigation. 3. Third-party control over payments. The moral hazard in physicians’ control noted in paragraph 1 above shows itself in those insurance schemes where the physician has the greatest control, namely, major medical insurance. Here there has been a marked rise in expenditures over time. In prepayment plans, where the insurance and medical service are supplied Q

111

Q

ARROW (1963)

by the same group, the incentive to keep medical costs to a minimum is strongest. In plans of the Blue Cross group, there has developed a conflict of interest between the insurance carrier and the medical-service supplier, in this case particularly the hospital. The need for third-party control is reinforced by another aspect of the moral hazard. Insurance removes the incentive on the part of individuals, patients, and physicians to shop around for better prices for hospitalization and surgical care. The market forces, therefore, tend to be replaced by direct institutional control. 4. Administrative costs. The pure theory of insurance sketched in Section B above omits one very important consideration: the costs of operating an insurance company. There are several types of operating costs, but one of the most important categories includes commissions and acquisition costs, selling costs in usual economic terminology. Not only does this mean that insurance policies must be sold for considerably more than their actuarial value, but it also means there is a great differential among different types of insurance. It is very striking to observe that among health insurance policies of insurance companies in 1958, expenses of one sort or another constitute 51.6 per cent of total premium income for individual policies, and only 9.5 per cent for group policies [26, Table 14-1, p. 272]. This striking differential would seem to imply enormous economies of scale in the provision of insurance, quite apart from the coverage of the risks themselves. Obviously, this provides a very strong argument for widespread plans, including, in particular, compulsory ones. Q

112

Q

ARROW (1963)

5. Predictability and insurance. Clearly, from the riskaversion point of view, insurance is more valuable, the greater the uncertainty in the risk being insured against. This is usually used as an argument for putting greater emphasis on insurance against hospitalization and surgery than other forms of medical care. The empirical assumption has been challenged by O. W. Anderson and others [3, pp. 53–54], who asserted that out-of-hospital expenses were equally as unpredictable as in-hospital costs. What was in fact shown was that the probability of costs exceeding $200 is about the same for the two categories, but this is not, of course, a correct measure of predictability, and a quick glance at the supporting evidence shows that in relation to the average cost the variability is much lower for ordinary medical expenses. Thus, for the city of Birmingham, the mean expenditure on surgery was $7, as opposed to $20 for other medical expenses, but of those who paid something for surgery the average bill was $99, as against $36 for those with some ordinary medical cost. Eighty-two per cent of those interviewed had no surgery, and only 20 per cent had no ordinary medical expenses [3, Tables A-13, A-18, and A-19 on pp. 72, 77, and 79, respectively]. The issue of predictability also has bearing on the merits of insurance against chronic illness or maternity. On a lifetime insurance basis, insurance against chronic illness makes sense, since this is both highly unpredictable and highly significant in costs. Among people who already have chronic illness, or symptoms which reliably indicate it, insurance in the strict sense is probably pointless. 6. Pooling of unequal risks. Hypothetically, insurance requires for its full social benefit a maximum possible discrimination Q

113

Q

ARROW (1963)

of risks. Those in groups of higher incidences of illness should pay higher premiums. In fact, however, there is a tendency to equalize, rather than to differentiate, premiums, especially in the Blue Cross and similar widespread schemes. This constitutes, in effect, a redistribution of income from those with a low propensity to illness to those with a high propensity. The equalization, of course, could not in fact be carried through if the market were genuinely competitive. Under those circumsances, insurance plans could arise which charged lower premiums to preferred risks and draw them off, leaving the plan which does not discriminate among risks with only an adverse selection of them. As we have already seen in the case of income redistribution, some of this may be thought of as insurance with a longer time perspective. If a plan guarantees to everybody a premium that corresponds to total experience but not to experience as it might be segregated by smaller subgroups, everybody is, in effect, insured against a change in his basic state of health which would lead to a reclassification. This corresponds precisely to the use of a level premium in life insurance instead of a premium varying by age, as would be the case for term insurance. 7. Gaps and coverage. We may briefly note that, at any rate to date, insurances against the cost of medical care are far from universal. Certain groups—the unemployed, the institutionalized, and the aged—are almost completely uncovered. Of total expenditures, between one-fifth and one-fourth are covered by insurance. It should be noted, however, that over half of all hospital expenses and about 35 per cent of the medical payments of those with bills of $1,000 a year and Q

114

Q

ARROW (1963)

over, are included [26, p. 376]. Thus, the coverage on the more variable parts of medical expenditure is somewhat better than the over-all figures would indicate, but it must be assumed that the insurance mechanism is still very far from achieving the full coverage of which it is capable.

D. UNCERTAINTY OF EFFECTS OF TREATMENT 1. There are really two major aspects of uncertainty for an individual already suffering from an illness. He is uncertain about the effectiveness of medical treatment, and his uncertainty may be quite different from that of his physician, based on the presumably quite different medical knowledges. 2. Ideal insurance. This will necessarily involve insurance against a failure to benefit from medical care, whether through recovery, relief of pain, or arrest of further deterioration. One form would be a system in which the payment to the physician is made in accordance with the degree of benefit. Since this would involve transferring the risks from the patient to the physician, who might certainly have an aversion to bearing them, there is room for insurance carriers to pool the risks, either by contract with physicians or by contract with the potential patients. Under ideal insurance, medical care will always be undertaken in any case in which the expected utility, taking account of the probabilities, exceeds the expected medical cost. This prescription would lead to an economic optimum. If we think of the failure to recover mainly in terms of lost working time, then this policy would, in fact, maximize economic welfare as ordinarily measured. Q

115

Q

ARROW (1963)

3. The concepts of trust and delegation. In the absence of ideal insurance, there arise institutions which offer some sort of substitute guarantees. Under ideal insurance the patient would actually have no concern with the informational inequality between himself and the physician, since he would only be paying by results anyway, and his utility position would in fact be thoroughly guaranteed. In its absence he wants to have some guarantee that at least the physician is using his knowledge to the best advantage. This leads to the setting up of a relationship of trust and confidence, one which the physician has a social obligation to live up to. Since the patient does not, at least in his belief, know as much as the physician, he cannot completely enforce standards of care. In part, he replaces direct observation by generalized belief in the ability of the physician.36 To put it another way, the social obligation for best practice is part of the commodity the physician sells, even though it is a part that is not subject to thorough inspection by the buyer. One consequence of such trust relations is that the physician cannot act, or at least appear to act, as if he is maximizing his income at every moment of time. As a signal to the buyer of his intentions to act as thoroughly in the buyer’s behalf as possible, the physician avoids the obvious stigmata of profitmaximizing. Purely arms-length bargaining behavior would be incompatible, not logically, but surely psychologically, with the trust relations. From these special relations come the various forms of ethical behavior discussed above, and so also, I suggest, the relative unimportance of profit-making in hospitals. The very word, “profit,” is a signal that denies the trust relations. Q

116

Q

ARROW (1963)

Price discrimination and its extreme, free treatment for the indigent, also follow. If the obligation of the physician is understood to be first of all to the welfare of the patient, then in particular it takes precedence over financial difficulties. As a second consequence of informational inequality between physician and patient and the lack of insurance of a suitable type, the patient must delegate to the physician much of his freedom of choice. He does not have the knowledge to make decisions on treatment, referral, or hospitalization. To justify this delegation, the physician finds himself somewhat limited, just as any agent would in similar circumstances. The safest course to take to avoid not being a true agent is to give the socially prescribed “best” treatment of the day. Compromise in quality, even for the purpose of saving the patient money, is to risk an imputation of failure to live up to the social bond. The special trust relation of physicians (and allied occuptions, such as priests) extends to third parties so that the certifications of physicians as to illness and injury are accepted as especially reliable (see Section II.B above). The social value to all concerned of such presumptively reliable sources of information is obvious. Notice the general principle here. Because there are barriers to the information flow and because there is no market in which the risks involved can be insured, coordination of purchase and sales must take place through convergent expectations, but these are greatly assisted by having clear and prominent signals, and these, in turn, force patterns of behavior which are not in themselves logical necessities for optimality.37 Q

117

Q

ARROW (1963)

4. Licensing and educational standards. Delegation and trust are the social institutions designed to obviate the problem of informational inequality. The general uncertainty about the prospects of medical treatment is socially handled by rigid entry requirements. These are designed to reduce the uncertainty in the mind of the consumer as to the quality of product insofar as this is possible.38 I think this explanation, which is perhaps the naive one, is much more tenable than any idea of a monopoly seeking to increase incomes. No doubt restriction on entry is desirable from the point of view of the existing physicians, but the public pressure needed to achieve the restriction must come from deeper causes. The social demand for guaranteed quality can be met in more than one way, however. At least three attitudes can be taken by the state or other social institutions toward entry into an occupation or toward the production of commodities in general; examples of all three types exist. (1) The occupation can be licensed, nonqualified entrants being simply excluded. The licensing may be more complex than it is in medicine; individuals could be licensed for some, but not all, medical activities, for example. Indeed, the present all-ornone approach could be criticized as being insufficient with regard to complicated specialist treatment, as well as excessive with regard to minor medical skills. Graded licensing may, however, be much harder to enforce. Controls could be exercised analogous to those for foods; they can be excluded as being dangerous, or they can be permitted for animals but not for humans. (2) The state or other agency can certify or label, without compulsory exclusion. The category of Certified Psychologist is now under active discussion; canned Q

118

Q

ARROW (1963)

goods are graded. Certification can be done by nongovernmental agencies, as in the medical-board examinations for specialists. (3) Nothing at all may be done; consumers make their own choices. The choice among these alternatives in any given case depends on the degree of difficulty consumers have in making the choice unaided, and on the consequences of errors of judgment. It is the general social consensus, clearly, that the laissez-faire solution for medicine is intolerable. The certification proposal never seems to have been discussed seriously. It is beyond the scope of this paper to discuss these proposals in detail. I wish simply to point out that they should be judged in terms of the ability to relieve the uncertainty of the patient in regard to the quality of the commodity he is purchasing, and that entry restrictions are the consequences of an apparent inability to devise a system in which the risks of gaps in medical knowledge and skill are borne primarily by the patient, not the physician.

POSTSCRIPT I wish to repeat here what has been suggested above in several places: that the failure of the market to insure against uncertainties has created many social institutions in which the usual assumptions of the market are to some extent contradicted. The medical profession is only one example, though in many respects an extreme one. All professions share some of the same properties. The economic importance of personal and especially family relationships, though declining, is by no means trivial in the most advanced economies; it Q

119

Q

ARROW (1963)

is based on non-market relations that create guarantees of behavior which would otherwise be afflicted with excessive uncertainty. Many other examples can be given. The logic and limitations of ideal competitive behavior under uncertainty force us to recognize the incomplete description of reality supplied by the impersonal price system.

NOTES * The author is professor of economics at Stanford University. He wishes to express his thanks for useful comments to F. Bator, R. Dorfman, V. Fuchs, Dr. S. Gilson, R. Kessel, S. Mushkin, and C. R. Rorem. This paper was prepared under the sponsorship of the Ford Foundation as part of a series of papers on the economics of health, education, and welfare. 1. This point has been stressed by I. M. D. Little [19, pp. 71–74]. For the concept of a “persuasive definition,” see C. L. Stevenson [27, pp. 210–17]. 2. The separation between allocation and distribution even under the above assumptions has glossed over problems in the execution of any desired redistribution policy; in practice, it is virtually impossible to find a set of taxes and subsidies that will not have an adverse effect on the achievement of an optimal state. But this discussion would take us even further afield than we have already gone. 3. The basic theorems of welfare economics alluded to so briefly above have been the subject of voluminous literature, but no thoroughly satisfactory statement covering both the theorems themselves and the significance of exceptions to them exists. The positive assertions of welfare economics and their relation to the theory of competitive equilibrium are admirably covered in Koopmans [18], The best summary of the various ways in which the theorems can fail to hold is probably Bator’s [6].

Q

120

Q

ARROW (1963)

4. There are further minor conditions, for which see Koopmans [18, pp. 50–55]. 5. “For a more precise statement of the existence conditions, see Koopmans [18, pp. 56–60] or Debreu [12, Ch. 5]. 6. The theory, in variant forms, seems to have been first worked out by Allais [2], Arrow [5], and Baudier [7]. For further generalization, see Debreu [11] and [12, Ch. 7]. 7. It should also be remarked that in the presence of uncertainty, indivisibilities that are sufficiently small to create little difficulty for the existence and viability of competitive equilibrium may nevertheless give rise to a considerable range of increasing returns because of the operation of the law of large numbers. Since most objects of insurance (lives, fire hazards, etc.) have some element of indivisibility, insurance companies have to be above a certain size. But it is not clear that this effect is sufficiently great to create serious obstacles to the existence and viability of competitive equilibrium in practice. 8. One form of production of information is research. Not only does the product have unconventional aspects as a commodity, but it is also subject to increasing returns in use, since new ideas, once developed, can be used over and over without being consumed, and to difficulties of market control, since the cost of reproduction is usually much less than that of production. Hence, it is not surprising that a free enterprise economy will tend to underinvest in research; see Nelson [21] and Arrow [4]. 9. An important current situation in which normal market relations have had to be greatly modified in the presence of great risks is the production and procurement of modern weapons; see Peck and Scherer [23, pp. 581–82] (I am indebted for this reference to V. Fuchs) and [1, pp. 71–75]. 10. For an explicit statement of this view, see Baumol [8], But I believe this position is implicit in most discussions of the functions of government.

Q

121

Q

ARROW (1963)

11. Since writing the above, I find that Buchanan and Tullock [10, Ch. 13] have argued that all redistribution can be interpreted as “income insurance.” 12. For an illuminating survey to which I am much indebted, see S. Mushkin [20]. 13. In governmental demand, military power is an example of a service used only irregularly and unpredictably. Here too, special institutional and professional relations have emerged, though the precise social structure is different for reasons that are not hard to analyze. 14. Even with material commodities, testing is never so adequate that all elements of implicit trust can be eliminated. Of course, over the long run, experience with the quality of product of a given seller provides a check on the possibility of trust. 15. See [22, p. 463], The whole of [22, Ch. 10] is a most illuminating analysis of the social role of medical practice; though Parsons’ interest lies in different areas from mine, I must acknowledge here my indebtedness to his work. 16. I am indebted to Herbert Klarman of Johns Hopkins University for some of the points discussed in this and the following paragraph. 17. The belief that the ethics of medicine demands treatment independent of the patient’s ability to pay is strongly ingrained. Such a perceptive observer as René Dubos has made the remark that the high cost of anticoagulants restricts their use and may contradict classical medical ethics, as though this were an unprecedented phenomenon. See [13, p. 419], “A time may come when medical ethics will have to be considered in the harsh light of economics” (emphasis added). Of course, this expectation amounts to ignoring the scarcity of medical resources; one has only to have been poor to realize the error. We may confidently assume that price and income do have some consequences for medical expenditures.

Q

122

Q

ARROW (1963)

18. A needed piece of research is a study of the exact nature of the variations of medical care received and medical care paid for as income rises. (The relevant income concept also needs study.) For this purpose, some disaggregation is needed; differences in hospital care which are essentially matters of comfort should, in the above view, be much more responsive to income than, e.g., drugs. 19. This role is enhanced in a socialist society, where the state itself is actively concerned with illness in relation to work; see Field [14, Ch. 9]. 20. About 3 per cent of beds were in proprietary hospitals in 1958, against 30 per cent in voluntary nonprofit, and the remainder in federal, state, and local hospitals; see [26, Chart 4-2, p. 60]. 21. C. R. Rorem has pointed out to me some further factors in this analysis. (1) Given the social intention of helping all patients without regard to immediate ability to pay, economies of scale would dictate a predominance of community-sponsored hospitals. (2) Some proprietary hospitals will tend to control total costs to the patient more closely, including the fees of physicians, who will therefore tend to prefer community-sponsored hospitals. 22. Without trying to assess the present situation, it is clear in retrospect that at some point in the past the actual differential knowledge possessed by physicians may not have been much. But from the economic point of view, it is the subjective belief of both parties, as manifested in their market behavior, that is relevant. 23. The degree of subsidy in different branches of professional education is worthy of a major research effort. 24. Strictly speaking, there are four variables in the market for physicians: price, quality of entering students, quality of education, and quantity. The basic market forces, demand for medical services and supply of entering students, determine two relations among the four variables. Hence, if the nonmarket forces determine the last two, market forces will determine price and quality of entrants.

Q

123

Q

ARROW (1963)

25. The supply of Ph.D.’s is similarly governed, but there are other conditions in the market which are much different, especially on the demand side. 26. Today only the Soviet Union offers an alternative lower level of medical personnel, the feldshers, who practice primarily in the rural districts (the institution dates back to the 18th century). According to Field [14, pp. 98–100, 132–33], there is clear evidence of strain in the relations between physicians and feldshers, but it is not certain that the feldshers will gradually disappear as physicians grow in numbers. 27. The law does impose some limits on risk-shifting in contracts, for example, its general refusal to honor exculpatory clauses. 28. There may be an identification problem in this observation. If the failure of the market system is, or appears to be, greater in medical care than in, say, food an individual otherwise equally concerned about the two aspects of others’ welfare may prefer to help in the first. 29. See [16, pp. 118–37], The calculations involve many assumptions and must be regarded as tenuous; see the comments by C. Reinold Noyes in [16, pp. 407–10]. 30. It might be argued that the existence of racial discrimination in entrance has meant that some of the rejected applicants are superior to some accepted. However, there is no necessary connection between an increase in the number of entrants and a reduction in racial discrimination; so long as there is excess demand for entry, discrimination can continue unabated and new entrants will be inferior to those previously accepted. 31. One problem here is that the tax laws do not permit depreciation of professional education, so that there is a discrimination against this form of investment. 32. To anticipate later discussion, this condition is not necessarily fulfilled. When it comes to quality choices, the market may be inaccurate.

Q

124

Q

ARROW (1963)

33. It is assumed that there are two classes, rich and poor ; the price of medical services to the rich is twice that to the poor, medical expenditures by the rich are 20 per cent of those by the poor, and the elasticity of demand for medical services is .5 for both classes. Let us choose our quantity and monetary units so that the quantity of medical services consumed by the poor and the price they pay are both 1. Then the rich purchase .1 units of medical services at a price of 2. Given the assumption about the elasticities of demand, the demand function of the rich is DR(p) = .14 p–.5 and that of the poor is DP(p) = p–.5. The supply of medical services is assumed fixed and therefore must equal 1.1. If price discrimination were abolished, the equilibrium price, p, must satisfy the relation, DR ( p ) + DP ( p ) = 1.1,

and therefore p = 1.07. The quantities of medical care purchased by the rich and poor, respectively, would be DR ( p ) = .135 and DP ( p ) = .965.

The inverse demand functions, the price to be paid corresponding to any given quantity are d R (q) = .02/q 3 , and d P (q) = 1/q 2 . Therefore, the consumers’ surplus to the rich generated by the change is:

∫

.135

.1

(.02/q 2 ) dq − p(.135 − .1),

(1)

and similarly the loss in consumers’ surplus by the poor is:

∫

1

.965

(1/q 2 ) dq − p(1 − .965)

(2)

If (2) is subtracted from (1), the second terms cancel, and the aggregate increase in consumers’ surplus is .0156, or a little over 1 per cent of the initial expenditures.

Q

125

Q

ARROW (1963)

34. A striking illustration of the desire for security in medical care is provided by the expressed preferences of émigrés from the Soviet Union as between Soviet medical practice and German or American practice; see Field [14, Ch. 12], Those in Germany preferred the German system to the Soviet, but those in the United States preferred (in a ratio of 3 to 1) the Soviet system. The reasons given boil down to the certainty of medical care, independent of income or health fluctuations. 35. It is a popular belief that the Chinese, at one time, paid their physicians when well but not when sick. 36. Francis Bator points out to me that some protection can be achieved, at a price, by securing additional opinions. 37. The situation is very reminiscent of the crucial role of the focal point in Schelling’s theory of tacit games, in which two parties have to find a common course of action without being able to communicate; see [24, esp. pp. 225 ff.]. 38. How well they achieve this end is another matter. R. Kessel points out to me that they merely guarantee training, not continued good performance as medical technology changes.

REFERENCES 1. A. A. Alchian, K. J. Arrow, and W. M. Capron, An Economic Analysis of the Market for Scientists and Engineers, RAND RM2190-RC. Santa Monica 1958. 2. M. Allais, “Géneralisation des théories de l’équilibre économique général et du rendement social au cas du risque,” in Centre National de la Recherche Scientifique, Econometrie, Paris 1953, pp. 1–20.

Q

126

Q

ARROW (1963)

3. O. W. Anderson and Staff of the National Opinion Research Center, Voluntary Health Insurance in Two Cities. Cambridge, Mass. 1957. 4. K. J. Arrow, “Economic Welfare and the Allocation of Resources for Invention,” in Nat. Bur. Econ. Research, The Role and Direction of Inventive Activity: Economic and Social Factors, Princeton 1962, pp. 609–25. 5. ———, “Les rôle des valeurs boursières pour la répartition la meilleure des risques,” in Centre National de la Recherche Scientifique, Econometrie, Paris 1953, pp. 41–46. 6. F. M. Bator, “The Anatomy of Market Failure,” Quart. Jour. Econ. Aug. 1958, 72, 351–79. 7. E. Baudier, “L’introduction du temps dans la théorie de l’équilibre général,” Les Cahiers Economiques, Dec. 1959, 9–16. 8. W. J. Baumol, Welfare Economics and the Theory of the State. Cambridge, Mass. 1952. 9. K. Borch, “The Safety Loading of Reinsurance Premiums,” Skandinavisk Aktuariehdskrift, 1960, pp. 163–84. 10. J. M. Buchanan and G. Tullock, The Calculus of Consent. Ann Arbor 1962. 11. G. Debreu, “Une économique de l’incertain,” Economie Appliquée, 1960, 13, 111–16. 12. ———, Theory of Values. New York 1959. 13. R. Dubos, “Medical Utopias,” Daedalus, 1959, 88, 410–24. 14. M. G. Field, Doctor and Patient in Soviet Russia. Cambridge, Mass. 1957. 15. Milton Friedman, “The Methodology of Positive Economics,” in Essays in Positive Economics, Chicago 1953, pp. 3–43. 16. ——— and S. S. Kuznets, Income from Independent Professional Practice. Nat. Bur. Econ. Research, New York 1945. 17. R. A. Kessel, “Price Discrimination in Medicine,” Jour. Law and Econ., 1958, 1, 20–53. 18. T. C. Koopmans, “Allocation of Resources and the Price System,” in Three Essays on the State of Economic Science, New York 1957, pp. 1–120.

Q

127

Q

ARROW (1963)

19. I. M. D. Little, A Critique of Welfare Economics. Oxford 1950. 20. Selma Mushkin, “Towards a Definition of Health Economics,” Public Health Reports, 1958, 73, 785–93. 21. R. R. Nelson, “The Simple Economics of Basic Scientific Research,” Jour. Pol. Econ., June 1959, 67, 297–306. 22. T. Parsons, The Social System. Glencoe 1951. 23. M. J. Peck and F. M. Scherer, The Weapons Acquisition Process: An Economic Analysis. Div. of Research, Graduate School of Business, Harvard University, Boston 1962. 24. T. C. Schelling, The Strategy of Conflict. Cambridge, Mass. 1960. 25. A. K. Shapiro, “A Contribution to a History of the Placebo Effect,” Behavioral Science, 1960, 5, 109–35. 26. H. M. Somers and A. R. Somers, Doctors, Patients, and Health Insurance. The Brookings Institution, Washington 1961. 27. C. L. Stevenson, Ethics and Language. New Haven 1945. 28. U. S. Department of Health, Education and Welfare, Physicians for a Growing America, Public Health Service Publication No. 709, Oct. 1959.

APPENDIX ON OPTIMAL INSURANCE POLICIES The two propositions about the nature of optimal insurance policies asserted in Section IV.B above will be proved here. Proposition 1. If an insurance company is willing to offer any insurance policy against loss desired by the buyer at a premium which depends only on the policy’s actuarial value, then the policy chosen by a risk-averting buyer will take the form of 100 per cent coverage above a deductible minimum.

Q

128

Q

ARROW (1963)

Note: The premium will, in general, exceed the actuarial value; it is only required that two policies with the same actuarial value will be offered by the company for the same premium. Proof: Let W be the initial wealth of the individual, X his loss, a random variable I(X ) the amount of insurance paid if loss X occurs, P the premium, and Y (X ) the wealth of the individual after paying the premium, incurring the loss, and receiving the insurance benefit. Y(X) = W − P − X + I (X).

(1)

The individual values alternative policies by the expected utility of his final wealth position Y (X ). Let U (y) be the utility of final wealth y; then his aim is to maximize, E{U [Y (X)]},

(2)

where the symbol E denotes mathematical expectation. An insurance payment is necessarily nonnegative, so the insurance policy must satisfy the condition, I (X ) ≥ 0 for all X.

(3)

If a policy is optimal, it must in particular be better in the sense of the criterion (2), than any other policy with the same actuarial expectation, E [I (X )]. Consider a policy that pays some positive amount of insurance at one level of loss, say X1, but which permits the final wealth at some other loss level, say X2, to be lower than that corresponding to X1.

Q

129

Q

ARROW (1963)

Then it is intuitively obvious that a risk-averter would prefer an alternative policy with the same actuarial value which would offer slightly less protection for losses in the neighborhood of X1 and slightly higher protection for those in the neighborhood of X2, since risk aversion implies that the marginal utility of Y (X ) is greater when Y (X ) is smaller: hence, the original policy cannot be optimal. To prove this formally, let I1(X ) be the original policy, with I1(X ) > 0 and Y1(X1) > Y2(X2), where Y1(X ) is defined in terms of I1(X ) by (I ). Choose d sufficiently small so that, I1(X ) > 0 for X1 ≤ X ≤ X1 + d ,

(4)

Y1(X’ ) < Y1(X ) for X2 X' ≤ X2 + d , X1 ≤ X ≤ X1 + d . (5)

(This choice of d is possible if the functions I1(X ), Y1(X ) are continuous; this can be proved to be true for the optimal policy, and therefore we need only consider this case.) Let p1 be the probability that the loss, X, lies in the interval 〈 X 1 , X 1 + δ 〉 and π 2 be the probability that X lies in the interval 〈 X 2 , X 2 + δ 〉. From (4) and (5) we can choose ⑀ > 0 and sufficiently small so that, I 1 ( X ) − π 2⑀ ≥ 0 for X 1 ≤ X ≤ X 1 + δ ,

(6)

Y1 ( X ′) + π 1⑀ < Y1 ( X ) − π 2⑀

(7)

for

X 2 ≤ X ′ ≤ X 2 + δ , X1 ≤ X ≤ X1 + δ .

Q

130

Q

ARROW (1963)

Now define a new insurance policy, I2(X ), which is the same as I1(X ) except that it is smaller by π 2⑀ in the interval from X1 to X1 + d and larger by π 1⑀ in the interval from X2 to X2 + d . From (6), I 2 ( X ) ≥ 0 everywhere, so that (3) is satisfied. We will show that E[ I 1 ( X )] = E[ I 2 ( X )] and that I2(X ) yields the higher expected utility, so that I1(X ) is not optimal. Note that I2(X ) – I1(X ) equals – π 2⑀ for X 1 ≤ X ≤ X1 + δ , π 1⑀ for X 2 ≤ X ≤ X 2 + δ , and 0 elsewhere. Let φ ( X ) be the density of the random variable X. Then E[ I 2 ( X ) − I 1 ( X )] = ∫

X1 + δ X1

+∫

[ I 2 ( X ) − I 1 ( X )]φ ( X ) dX

X2 +δ X2

= (−π 2⑀ ) ∫

[ I 2 ( X ) − I 1 ( X )] dX X1 + δ X1

φ ( X ) dX + (π 1⑀ ) ∫

X2 +δ X2

φ ( X ) dX

= −(π 2⑀ )π 1 + (π 1⑀ )π 2 = 0,

so that the two policies have the same actuarial value and, by assumption, the same premium. Define Y2(X ) in terms of I2(X ) by (1). Then Y2(X ) – Y1(X ) = I2(X ) – I1(X ). From (7), Y1 ( X ′) < Y2 ( X ′) < Y2 ( X ) < Y1 ( X ) for

(8)

X 2 ≤ X ′ ≤ X 2 + δ , X1 ≤ X ≤ X1 + δ .

Since Y1 ( X ) − Y2 ( X ) = 0 outside the intervals 〈 X1 , X 1 + δ 〉, 〈 X 2 , X 2 + δ 〉, we can write, E {U [Y2 ( X )] − U [Y1 ( X )]} = ∫

+∫

X 2 +δ X2

X1 +δ X1

{U [Y2 ( X )] − U [Y1 ( X )]}φ ( X ) dX

{U [Y2 ( X )] − U [Y1 ( X )]}φ ( X ) dX .

Q

131

Q

(9)

ARROW (1963)

By the Mean Value Theorem, for any given value of X, U [Y2 ( X )] − U [Y1 ( X )] = U ′[Y ( X )][Y2 ( X ) − Y1 ( X )] = U ′[Y ( X )][ I 2 ( X ) − I 1 ( X )],

(10)

where Y (X ) lies between Y1(X ) and Y2(X ). From (8), Y ( X ′) < Y ( X ) for X 2 ≤ X ′ ≤ X 2 + δ , X 1 ≤ X ≤ X 1 + δ ,

and, since U' ( y) is a diminishing function of y for a risk-averter, U ′[Y ( X ′)] > U ′[Y ( X )]

or, equivalently, for some number u, U ′[Y ( X ′)] > u for X 2 ≤ X ′ ≤ X 2 + δ , U ′[Y ( X )] < u for X 1 ≤ X ≤ X 1 + δ .

(11)

Now substitute (10) into (9), E {U [Y2 ( X )] − U [Y1 ( X )]} = −π 2⑀ ∫

X1 +δ X1

X + π 1⑀ ∫

U ′[Y ( X )]φ ( X ) dX

X 2 +δ X2

U ′[Y ( X )]φ ( X ) dX .

From (11), it follows that, E {U [Y2 ( X )] − U [Y1 ( X )]} > −π 2⑀ uπ 1 + π 1⑀ uπ 2 = 0,

so that the second policy is preferred. It has thus been shown that a policy cannot be optimal if, for some X1 and X2, I(X1) > 0, Y(X1) > Y(X2). This may be put in a different form: Let Ymin be the minimum value taken Q

132

Q

ARROW (1963)

on by Y(X ) under the optimal policy; then we must have I(X ) = 0 if Y(X ) > Ymin. In other words, a minimum final wealth level is set; if the loss would not bring wealth below this level, no benefit is paid, but if it would, then the benefit is sufficient to bring up the final wealth position to the stipulated minimum. This is, of course, precisely a description of 100 per cent coverage for loss above a deductible. We turn to the second proposition. It is now supposed that the insurance company, as well as the insured, is a riskaverter; however, there are no administrative or other costs to be covered beyond protection against loss. Proposition 2. If the insured and the insurer are both riskaverters and there are no costs other than coverage of losses, then any nontrivial Pareto-optimal policy, I(X ), as a function of the loss, X, must have the property, 0 < dI /dX < 1. That is, any increment in loss will be partly but not wholly compensated by the insurance company; this type of provision is known as coinsurance. Proposition 2 is due to Borch [9, Sec. 2]; we give here a somewhat simpler proof. Proof: Let U( y) be the utility function of the insured, V (z) that of the insurer. Let W0 and W1 be the initial wealths of the two, respectively. In this case, we let I(X) be the insurance benefits less the premium; for the present purpose, this is the only significant magnitude (since the premium is independent of X, this definition does not change the value of dI/dX ). The final wealth positions of the insured and insurer are: Y ( X ) = W0 − X + I ( X ), Z ( X ) = W1 − I ( X ), Q

133

Q

(12)

ARROW (1963)

respectively. Any given insurance policy then defines expected utilities, u = E {U [Y ( X )]} and v = E {V [ Z ( X )]}, for the insured and insurer, respectively. If we plot all points (u, v) obtained by considering all possible insurance policies, the resulting expected-utility-possibility set has a boundary that is convex to the northeast. To see this, let I1(X ) and I2(X ) be any two policies, and let (u1, v1) and (u2, v2) be the corresponding points in the two-dimensional expectedutility-possibility set. Let a third insurance policy, I(X ), be defined as the average of the two given ones,

I ( X ) = ( 21 ) I 1 ( X ) + ( 21 ) I 2 ( X ), for each X. Then, if Y(X ), Y1(X ), and Y2(X ) are the final wealth positions of the insured, and Z(X ), Z1(X ), and Z2(X ) those of the insurer for each of the three policies, I(X ), I1(X ), and I2(X ), respectively,

Y ( X ) = ( 21 )Y1 ( X ) + ( 21 )Y2 ( X ), Z ( X ) = ( 21 )Z1 ( X ) + ( 21 ) Z2 ( X ), and, because both parties have diminishing marginal utility,

U [Y ( X )] ≥ ( 21 )U [Y1 ( X )] + ( 21 )U [Y2 ( X )], V [ Z ( X )] ≥ ( 21 )V [ Z1 ( X )] + ( 21 )V [ Z2 ( X )]. Since these statements hold for all X, they also hold when expectations are taken. Hence, there is a point

Q

134

Q

ARROW (1963)

(u, v) in the expected-utility-possibility set for which u ≥ ( 21 ) u1 + ( 21 )u2 , v ≥ ( 21 )v1 + ( 21 )v2 . Since this statement holds for every pair of points (u1, y1) and (u2, v2) in the expectedutility-possibility set, and in particular for pairs of points on the northeast boundary, it follows that the boundary must be convex to the northeast. From this, in turn, it follows that any given Pareto-optimal point (i.e., any point on the northeast boundary) can be obtained by maximizing a linear function, a u + b v, with suitably chosen a and b nonnegative and at least one positive, over the expected-utility-possibility set. In other words, a Paretooptimal insurance policy, I(X ), is one which maximizes, α E {U [Y ( X )]} + β E {V [ Z ( X )]} = E {αU [Y ( X )] + βV [ Z ( X )]},

for some α ≥ 0, β ≥ 0, α > 0 or β > 0. To maximize this expectation, it is obviously sufficient to maximize: αU [Y ( X )] + βV [ Z ( X )],

(13)

with respect to I(X ), for each X. Since, for given X, it follows from (12) that, dY ( X )/dI ( X ) = 1, dZ ( X )/dI ( X ) = −1,

it follows by differentiation of (13) that I(X ) is the solution of the equation, αU ′[Y ( X )] − βV ′[ Z ( X )] = 0.

The cases a = 0 or b = 0 lead to obvious trivialities (one party simply hands over all his wealth to the other), so we

Q

135

Q

ARROW (1963)

assume a > 0, b > 0. Now differentiate (14) with respect to X and use the relations, derived from (12), dY /dX = (dI /dX ) − 1, dZ /dX = −(dI /dX ).

αU ′′[Y ( X )][ dI /dX ) − 1] + βV ′′[ Z ( X )](dI /dX ) = 0,

or dI /dX = αU ′′[Y ( X )]/ {αU ′′[Y ( X )] + βV ′′[ Z ( X )]}.

Since U '' [Y(X )] < 0, V'' [Z(X )] < 0 by the hypothesis that both parties are risk-averters, Proposition 2 follows.

Q

136

Q

NOTES ON CONTRIBUTORS

is a Nobel laureate in economic sciences and professor emeritus at Stanford University. He was awarded the Nobel Prize in 1972 “for pioneering contributions to general economic equilibrium theory and welfare theory.” Arrow began his graduate study in economics and statistics at Columbia University, earning his PhD there. He has held positions on the Cowles Commission for Research in Economics, at the University of Chicago, at Harvard University, and at Stanford University. His research, apart from social choice theory, has focused on general economic equilibrium. The profound transformation of the general equilibrium theory is marked by his groundbreaking work. He helped open new productive paths for research in this area and in so doing has made fundamental contributions to the renewal of the theory. KENNETH J. ARROW

is the Ford Professor of Economics at the Massachusetts Institute of Technology. She is also the codirector of the Public Economics Program at the National Bureau of Economic Research, where she is a research associate. She is a co-editor of the Journal of Public Economics, a member of the Institute of Medicine and of the American Academy of Arts and Sciences, and a fellow of the Econometric Society. Finkelstein received her AB summa cum laude in government from Harvard University in 1995 and an MPhil AMY FINKELSTEIN

NOTES ON CONTRIBUTORS

in economics from Oxford University in 1997, where she was a Marshall Scholar. She received her PhD in economics from MIT in 2001. Prior to joining the MIT faculty in 2005, she was a junior fellow at the Harvard Society of Fellows. She has received numerous awards and fellowships including the John Bates Clark Medal (2012), a Presidential Early Career Award for Scientists and Engineers (2009), the American Economic Association’s Elaine Bennett Research Prize (2008), and a Sloan Research Fellowship (2007), as well as awards for graduate student teaching and graduate student advising at MIT. is a professor of economics at the Massachusetts Institute of Technology. He is also the director of the Health Care Program at the National Bureau of Economic Research, where he is a research associate. He is an associate editor of both the Journal of Public Economics and the Journal of Health Economics. In 2009, he was elected to the Executive Committee of the American Economic Association. He is also a member of the Institute of  Medicine, the American Academy of Arts and Sciences, and the National Academy of Social Insurance. He received his BS in economics from MIT and his PhD in economics from Harvard University. He has published more than 140 research articles, has edited six research volumes, and is the author of Public Finance and Public Policy, a leading undergraduate text, and Health Care Reform, a graphic novel. In 2006, he received the American Society of Health Economists Inaugural Medal for the best health economist in the nation aged 40 and under. JONATHAN GRUBER

is the John D. MacArthur Professor of Health Policy and Management at Harvard University, where he is a member of the faculties of the Harvard Kennedy School, the Harvard Medical School, the Harvard School of Public Health, and the Faculty of Arts and Sciences. He is also a Faculty Research Associate of the National Bureau of Economic Research. He received his BA and PhD in economics from Harvard University in 1963 and 1969, respectively. He was the founding editor of the Journal of Health Economics,

JOSEPH P. NEWHOUSE

Q

138

Q

NOTES ON CONTRIBUTORS

which he edited for 30 years, and is a member of the Institute of Medicine and a fellow of the American Academy of Arts and Sciences. In 2014 he won the Victor Fuchs Lifetime Achievement Award from the American Society of Health Economists, and he was the first recipient of the David N. Kershaw Award and Prize of the Association for Public Policy and Management in 1983. In 2001 and again in 2013 he and his co-authors won the Kenneth J. Arrow Award for the best paper in health economics for How Does Managed Care Do It? and The Oregon Health Insurance Experiment; the latter also won the 2013 HSR Impact Award from AcademyHealth. received his PhD from MIT in 1967, became a full professor at Yale in 1970, and in 1979 was awarded the John Bates Clark Award. He has taught at Princeton, Stanford, and MIT and was the Drummond Professor and a fellow of All Souls College, Oxford. He is now University Professor at Columbia University in New York and a member (and former chair) of Columbia University’s Committee on Global Thought. He is also the founder and copresident of the Initiative for Policy Dialogue at Columbia. In 2001, he was awarded the Nobel Prize in economics for his analyses of markets with asymmetric information, and he was a lead author of the 1995 Report of the Intergovernmental Panel on Climate Change, which shared the 2007 Nobel Peace Prize. In 2011, Time named Stiglitz one of the 100 most influential people in the world.

JOSEPH E. STIGLITZ

Q

139

Q

INDEX

ability to pay: in hospitals, 123n21; medical ethics and, 122n17 accountability, in health care, 63 Acemoglu, Daron, 39 administered prices, 99 administrative costs, of health insurance, 112 adverse selection: Arrow on, 58–59; health insurance and, 17; highdeductible health insurance and, 31–32; risk pooling and, 114 advertising, 61; physicians and, 93 affordability: cost-sharing and, 50–51; of deductible, 51; of health insurance, 50–51; out-of-pocket maximum and, 51 Affordable Care Act (Obamacare), ix, 68; coverage expansion with, 37; lifetime limits and, 10n5; uninsured and, 14 aging population, health care spending and, 36 Akerlof, George, 2–3 alcohol, 80 American Medical Association, 97 Anderson, O. W., 113 Aron-Dine, Aviva, 28 Arrow, Kenneth J., x, 1–4, 7, 13, 14–15, 24, 40, 46, 52, 66; on adverse selection, 58–59; “Aspects of the Theory of Risk Bearing” by, 65, 69;

on preventative care, 76–79; Social Choice and Individual Values by, 56; “Uncertainty and the Welfare Economics of Medical Care” by, 1, 13, 46, 81–120 Arrow-Debreu securities, viii “Aspects of the Theory of Risk Bearing” (Arrow), 65, 69 asymmetric information, 56; between doctor and patient, 58, 69; in health care, 60, 66–70, 96; health insurance and, 60, 69–70; ideal health insurance and, 116; uncertainty and, 96 Baicker, Kate, 18 Becker, Gary, 15 Bernoulli, Daniel, 106–7 Blue Cross: closed-panel practice and, 104; conflict of interest with, 112; risk pooling by, 114 capitalism, 62 capitation, 63; in HMOs, 80 catastrophic coverage: high-deductible health insurance and, 25; in RAND Health Insurance Experiment, 23 certification, 119 charity treatment, 93–94; price discrimination and, 104

INDEX

employer-provided health insurance, 28–29; end-of-year price and, 28; health care spending and, 30; highdeductible health insurance and, 25–26; marginal utility and, 3–4; with prepayment, 111; risk aversion and, 128; spot price and, 28–29. See also high-deductible health insurance delegation, to physician, 117 demand: elasticity of, price discrimination and, 103, 104–5; in health care market, 92; purchasing power transfer and, 83–84. See also price sensitivity of demand demand curve, 4; deadweight loss and, 6–7; expected negative value and, 8; ex post moral hazard and, 16; price elasticity and, 5, 6n10, 16 The Demand for Health: A Theoretical and Empirical Investigation (Grossman), 1 diagnosis: Medicare payments based on, 48; risk of, 57–58 doughnut hole, with Medicare Part D, 41 drugs: compliance with, 8; control mechanisms for use of, 74; Medicare and, 39, 41, 68; price negotiation for, 68 Dubos, Renê, 122n14

Chetty, Raj, 50 chronic illness, predictability of, 113 closed-panel practice, 98–99; Blue Cross and, 104; prepayment and, 104 co-insurance: co-payments and, 52; with deductible, 4; incentives and, 4; moral hazards and, 4; with prepayment, 111; price elasticity and, 6; for risk aversion, 110; risks with, 4; uncertainty and, 50–51 collectivity-orientation, 93, 94–95 commodities: information as, 88; quality control with, 122n14 communicable diseases, 86, 99 competitive equilibrium, 2; existence of, 85–86; marketability and, 85–87; nonincreasing returns in, 85; optimality and, 82–85; risk and, 87–91; welfare economics and, 120n3 competitive model: entry restrictions and, 101–3; for health care, 82, 85, 99–105; ideal, 105–20; increasing returns in, 100; price discrimination in, 103–5; reality and, 85; uncertainty and, 99–105. See also ideal competitive model compliance, with drugs, 8 conflict of interest, with Blue Cross, 112 Congressional Budget Office, 30 contract reimbursement systems, 63, 98–99 co-payments: co-insurance and, 52; drug compliance and, 8; provider-side moral hazard and, 48 corporate culture, in health care, 62–63 cost-price effects, 8 cost-sharing: affordability and, 50–51; efficacious services and, 7; out-ofpocket price and, 5; in RAND Health Insurance Experiment, 22–24 Cullen, Mark, 28, 31

Ehrlich, Isaac, 15 Einav, Liran, 28, 31 émigrés, in Soviet Union, 126n34 employer-provided health insurance: deductible in, 28–29; selection on moral hazard and, 32–33 end-of-year price, 28 entry restrictions: competitive model and, 101–3; racial discrimination and, 124n30 Evans, Robert, 6 ex ante moral hazard, 15–16; Oregon Health Insurance Experiment and, 21–22, 47; unimportance of, 47 expected negative value, demand curve and, 8

Dasgupta, Partha, viii deadweight loss, 4; demand curve and, 6–7; from moral hazards, 8 deductible: affordability of, 51; co-insurance with, 4; in

Q

141

Q

INDEX

high-deductible health insurance and, 33–34; income distribution and, 123n18; inefficiency of, 67; inferiority of U.S. system, 67; international differences in, 62; liquidity effect and, 49–50; marginal price and, 27; out-of-pocket cost and, 27; out-ofpocket price for, 3; price sensitivity of demand for, 15–18; price transparency of, 52; RAND Health Insurance Experiment and, 30–31; rent-seeking in, 67–68; risk and, 89, 92, 105–6; in Soviet Union, 94; taxes and, 78; technology and, 39; uncertainty in, 89, 115–19; utility in, 106–7; value judgments on, 83–84. See also preventative care health care market, 91–99; demand in, 92; licensing and, 96; medical education and, 96–97; physicians in, 93–95, 123n24; price in, 98–99; profit motive in, 90; quality control in, 98; risks and, 2; selection in, 2–3; supply in, 96–98; uncertainty in, 95–96. See also market failure health care spending: GDP and, 14, 35, 35, 77–78; hospital admissions and, 75; income distribution and, 79; National Institutes of Health and, 77; out-of-pocket costs and, 26–28 health insurance: administrative costs of, 112; adverse selection and, 17; affordability of, 50–51; alternative payment methods for, 110–11; asymmetric information and, 60, 69–70; gaps in coverage in, 114–15; health care increases from, 14; health care spending and, 36, 38; ideal, 106–9, 115; incentives and, 109–10, 112; liquidity effect and, 49–50; medical care insurance and, 2n9–10; mortality and, 18; optimality in, 128– 36; predictability and, 113; premium loading in, 108; risk and, 57–58; risk aversion and, 107–9; risk pooling by, 107–8, 113–14; substitution effect and, 49, 50; technology and, 36, 41;

expected utility hypothesis, 106–7 ex post moral hazard, 16, 47 fee-for-provider, 79 fee-for-service system, 63; Medicare, 48 feldshers, in Soviet Union, 124n26 Feldstein, Martin, 51 Finkelstein, Amy, viii, ix–x, 3, 45–47, 55–56; on preventative care, 74–76 First Optimality Theorem, 82 Flexner Report, 97 Ford Foundation, 56 free-care plan, in RAND Health Insurance Experiment, 23 Free for All (Newhouse), 22 Friedman, Milton, 61, 85, 101 Fuchs, Victor, 56 future prices, with employer-provided health insurance, 29 future spending, preventative care and, 74–80 gaps in coverage, in health insurance, 114–15 GDP, health care spending and, 14, 35, 35, 66, 77–78 general equilibrium theory, 1–2; provider-side moral hazard and, 47 Gladwell, Malcolm, 16–17, 18 global warming, viii government: market failure and, 90; need for, 99; taxes and, 79 Greenwald, Bruce, 65–66 Grossman, Michael, 1 Gruber, Jonathan, x, 55–56 Hall, Robert, 77 health care: accountability in, 63; aging population and, 36; asymmetric information in, 60, 66–70, 96; competitive model for, 82, 85, 99–105; corporate culture in, 62–63; deductible and, 30; demand for, 15–18; ex post moral hazard and, 16; GDP and, 66; growth of, 34–40, 35; health insurance and, 36, 38; with high-deductible health insurance, 30;

Q

142

Q

INDEX

health insurance (continued ) third-party payments with, 111–12; transaction costs for, 68; uninsured and, 14, 79, 114–15; utility and, 133–35; welfare cost of, 41–42. See also employer-provided health insurance; high-deductible health insurance; ideal health insurance; Medicaid; Medicare; Oregon Health Insurance Experiment; RAND Health Insurance Experiment Health Savings Accounts Act of 2003, 25 heterogeneity, price sensitivity of demand and, 51 high-deductible health insurance, 25–26; adverse selection and, 31–32; catastrophic coverage and, 25; health care spending and, 30, 33–34; Health Savings Accounts Act of 2003 and, 25; price sensitivity of demand and, 31; selection of, 31–34; tax subsidies for, 33 Hippocrates, 61 hiring date, with employer-provided health insurance, 28–29 HMOs, 61–62; capitation in, 80 hospital admissions: health care spending and, 75; Medicaid and, 19–20, 20; Oregon Health Insurance Experiment and, 19–20, 20, 21 hospitalization insurance, 57 hospitals: ability to pay in, 123n21; incentives of, 39; increasing returns of, 100; length of stay in, 48; Medicare and, 37, 48; physicians and, 94; profit motive in, 116; spending by, 37 hyperbolic discounting, drug compliance and, 8 ideal competitive model: ideal health insurance and, 106–9, 115, 116; ideal insurance and, 106–9, 115, 116; problems of insurance and, 109–15; treatment effects and, 115–19; uncertainty and, 105–20 ideal health insurance, 106–9, 115, 116

Q

143

immunizations, 86 incentives, ix, 59; co-insurance and, 4; ex ante moral hazard and, 9; health insurance and, 109–10, 112; of hospitals, 39; of physicians, 9; with prepayment, 111–12; for technology, 36 income, utility and, 106 income distribution: health care and, 123n18; health care spending and, 79; heterogeneity and, 51; price and, 125n33; price discrimination and, 104–5; redistribution of, 84, 90–91, 104–5, 114; risk pooling and, 114 income effect, 49 incomplete risk markets, 66 increasing returns: in competitive model, 100; uncertainty and, 121n7 indemnities, 110, 111 individual costs, marketability and, 85 indivisibility, 100, 121n7 infant mortality, 67 information: risk and, 88. See also asymmetric information International Panel on Climate Change (IPCC), viii Jones, Charles, 77 Joskow, Paul, 45 Kaiser, 62 Kenneth J. Arrow Lecture, vii–xi, 13, 65 Kessel, R. A., 103 Kuznets, S. S., 101 licensing: of doctors, 61; entry restrictions and, 102–3; health care market and, 96; uncertainty and, 118–19 lifetime limits, Affordable Care Act and, 10n5 Linn, Josh, 39 liquidity effect: health care spending and, 49–50; health insurance and, 49–50; unemployment insurance and, 50 lottery, for Medicaid, 18–22

Q

INDEX

mortality: health insurance and, 18; infant, 67

major risk insurance, 51 marginal price, 27, 52 marginal utility, 3–4 marketability: competitive equilibrium and, 85–87; risk and, 87–91 market failure, 89, 124n28; adverse selection and, 59; government and, 90; marketability and, 87; optimality and, 90 Maskin, Eric, vii–viii maternity, predictability of, 113 McGarry, Kathleen, 46 Mean Value Theorem, 132 Medicaid: hospital admissions and, 19–20, 20; lottery for, 18–22; Oregon Health Insurance Experiment on, 18–22, 20; projected federal spending growth in, 14 medical care. See health care medical education: entry restrictions in, 101; health care market and, 96–97; quality control in, 118–19; racial discrimination in, 124n30; subsidies in, 96–97, 123n23; taxes and, 124n31 medical ethics: ability to pay and, 122n17; for monopolization, 60–61; physicians and, 93; price-fixing and, 99 Medicare: coverage expansion with, 36–38; diagnosis and, 48; drugs and, 39, 41, 68; hospital spending and, 37; hospital stays with, 48; introduction of, 36–38; new drug trials and, 39; Part D of, 41; projected federal spending growth in, 14; providerside moral hazard and, 48; technology and, 39–40 Medicare Advantage, 68 Medicare Fee Schedule, 9 monopolization: medical ethics for, 60–61; price discrimination and, 103 “The Moral-Hazard Myth” (Gladwell), 16–17 moral hazards, viii–ix; co-insurance and, 4; deadweight loss from, 8; Finkelstein and, 3. See also specific related topics; specific types

Q

National Institutes of Health, health care spending and, 77 new drug trials, Medicare and, 39 Newhouse, Joseph, x–xi, 22, 36 nonincreasing returns, in competitive equilibrium, 85 nonprofit hospitals, 94 nutrition, 80 Nyman, John, 17 Obamacare. See Affordable Care Act optimality: competitive equilibrium and, 82–85; in health insurance, 128–36; market failure and, 90; risk and, 88 Oregon Health Insurance Experiment: ex ante moral hazard and, 21–22, 47; hospital admissions and, 19–20, 20, 21, 75; on Medicaid, 18–22, 20; preventative care and, 76; prospective reimbursement system and, 49 out-of-pocket costs: health care spending and, 26–28, 27; with high-deductible health insurance, 25; in RAND Health Insurance Experiment, 23; uncertainty and, 52 out-of-pocket maximum: affordability and, 51; in RAND Health Insurance Experiment, 23 out-of-pocket price: cost-sharing and, 5; efficacious and nonefficacious medical services and, 7–8; for health care, 3 Pareto efficient, 66; optimality and, 83; premium loading and, 108 Parsons, Talcott, 93 Part D, of Medicare, 41 Pauly, Mark, 3, 6–7, 16 physicians: advertising and, 93; delegation to, 117; in health care market, 123n24; hospitals and, 94; incentives for, 9; income of, 101; medical ethics and, 93; profit motive of, 94; self-interest of, 93

144

Q

INDEX

purchasing power transfer: demand curve and, 83–84; taxes and, 84; tax subsidies and, 84

pooling, of risk, 107–8, 113–14 predictability, health insurance and, 113 premium loading, 108 prepayment, 98–99, 110–11; closed-panel practice and, 104; incentives with, 111–12 preventative care: Arrow on, 76–79; Finkelstein on, 74–76; future spending and, 74–80; Stiglitz on, 74–75, 79–80; tax subsidies for, 80 price: administered, 99; cost-price effects, 8; end-of-year, 28; future prices, with employer-provided health insurance, 29; in health care market, 98–99; income distribution and, 125n33; marginal, 27, 52; social cost and, 41; spot, 28–29; transparency of, of health care, 52. See also out-of-pocket price price discrimination: in competitive model, 103–5; trust and, 117 price elasticity: co-insurance and, 6; Congressional Budget Office and, 30; demand curve and, 5, 6n10, 16; for health care, 15–18; high-deductible health insurance and, 31; RAND Health Insurance Experiment and, 30–31; risk and, 89 price-fixing, 99 price sensitivity of demand: ex post moral hazard and, 16; heterogeneity and, 51; with major risk insurance, 51; selection on moral hazard and, 32, 33 profit maximization, 61; price discrimination and, 103–5 profit motive: in health care market, 90; in hospitals, 116; of physicians, 94; trust and, 116 prospective reimbursement system, 48, 49 provider-side moral hazard, 47–49, 75; general equilibrium theory and, 47; Medicare and, 48; patient-side moral hazard and, 48–49; with specialists, 73–74

Q

quality control: with commodities, 122n14; in health care market, 98; licensing and, 103; in medical education, 118–19 racial discrimination, 124n30 RAND Health Insurance Experiment, 3, 6–8; cost-sharing in, 22–24; health care spending and, 26, 30–31, 38; hospital admissions and, 75; out-of-pocket costs and, 26; price elasticity and, 30–31; prospective reimbursement system and, 49 Reinhardt, Uwe, 17 rent-seeking, 67–68 reputation mechanisms, 80 Reynolds, Lloyd, 56 risk: Arrow and, 57; with co-insurance, 4; competitive equilibrium and, 87–91; of diagnosis, 57–58; health care and, 89, 92, 105–6; health care markets and, 2; health insurance and, 57–58; information and, 88; marketability and, 87–91; optimality and, 88; price elasticity and, 89 risk aversion: co-insurance for, 110; deductible and, 128; health insurance and, 107–9, 128, 130, 132; uncertainty and, 113 risk pooling: by health insurance, 107–8, 113–14; income distribution and, 114 Rorem, C. Rufus, 57, 123n21 Rothschild, Michael, 2–3, 65 Ryan, Steve, 31 Samuelson, Paul, 45 Scheinkman, José, viii Schelling, T. C., 126n37 Schrimpf, Paul, 31 Second Optimality Theorem, 84 selection: of high-deductible health insurance, 31–34; in insurance markets, 2–3

145

Q

INDEX

selection on moral hazard, 31–33, 35 self-interest, of physicians, 93 Sen, Amartya, vii–viii smoking, 80 Social Choice and Individual Value (Arrow), 56 social costs: marketability and, 86; price and, 41 Solow, Robert, 45 Somers, Anne, 57 Somers, Harold, 57 Soviet Union: émigrés in, 126n34; feldshers in, 124n26; health care in, 94 specialists, provider-side moral hazard with, 73–74 spot price, 28–29 Stiglitz, Joseph, 2–3; on preventative care, 74–75, 79–80 subsidies: in medical education, 96–97, 123n23. See also tax subsidies substitution effect, health insurance and, 49, 50 supply, in health care market, 96–98 tacit games theory, 126n37 taxes, 120n3; government and, 79; health care and, 78; marketability and, 86; medical education and, 124n31; purchasing power transfer and, 84; welfare cost and, 79 tax subsidies, 120n3; for high-deductible health insurance, 33; marketability and, 86; for preventative care, 80; purchasing power transfer and, 84 technology: health care spending and, 39; health insurance and, 36, 41; incentives for, 36; marketability and, 87; Medicare and, 39–40; Stiglitz on, 74–75

Q

third-party payments, 111–12 transaction costs, for health insurance, 68 trust, 116 uncertainty, 2; adverse selection and, 59; Arrow and, 57; asymmetric information and, 96; co-insurance and, 50–51; competitive model and, 99–120; in health care, 89, 115–19; in health care market, 95–96; ideal competitive model and, 105–20; increasing returns and, 121n7; licensing and, 118–19; marginal price and, 52; out-of-pocket costs and, 52; risk aversion and, 113 “Uncertainty and the Welfare Economics of Medical Care” (Arrow), 1, 13, 46, 81–120 unemployment insurance, 50 unhealthy living, ex ante moral hazard and, 15–16 uninsured, 14, 79, 114–15 utility: in health care, 106–7; health insurance and, 133–35; income and, 106; marginal, 3–4 value judgments, on health care, 83–84 Weber, Max, 62 welfare cost: of health insurance, 41–42; taxes and, 79 welfare economics: competitive equilibrium and, 120n3; drug compliance and, 8. See also “Uncertainty and the Welfare Economics of Medical Care” Zeckhauser, Richard, 4

146

Q