<html><body>Is Everyone's Life Worth the Same?: Dilemmas for Regulators September 15, 2004 Unedited transcript prepared from a tape recording <TABLE width="100%" border=0> <TBODY> <TR> <TD vAlign=top align=left width="25%">11:45 a.m.</TD> <TD vAlign=top align=left width="75%" colSpan=2> Registration</TD></TR> <TR> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="50%"> </TD></TR> <TR> <TD vAlign=top align=left width="25%">Noon</TD> <TD vAlign=top align=left width="25%">Luncheon</TD> <TD vAlign=top align=left width="50%"> </TD></TR> <TR> <TD vAlign=top align=left width="25%">12:30 p.m.</TD> <TD vAlign=top align=left width="25%">Welcome:</TD> <TD vAlign=top align=left width="50%">Robert W. Hahn, AEI-Brookings Joint Center</TD></TR> <TR> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="25%">Panelists:</TD> <TD vAlign=top align=left width="50%">Cass Sunstein, University of Chicago Law School</TD></TR> <TR> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="50%">W. Kip Viscusi, Harvard University Law School</TD></TR> <TR> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="25%"> </TD> <TD vAlign=top align=left width="50%"> </TD></TR> <TR> <TD vAlign=top align=left width="25%">2:00</TD> <TD vAlign=top align=left width="75%" colSpan=2> Adjournment</TD></TR></TBODY></TABLE> Proceedings: MR. HAHN:  Welcome to Meditation 101.  Good afternoon. My name is Bog Hahn, and I am a scholar at AEI, and I also have the privilege of directing the AEI Brookings Joint Center. Professor Sunstein asked me while we were kibitzing over lunch are our turnouts always this large, and I suggested it was either because of the Sunstein-Viscusi effect or the free lunch, and we are going to ask Alan Krupnik [ph] to report to us after today's session. I want to start by thanking Sasha Gentling for putting this event together. You will note in your packet that you have a blue card, and if you want to be part of the Joint Center family and get our occasional emails, please fill out that card and give it to Sasha. We are going to have some upcoming conferences related to this event, including one on the so-called Copenhagen Consensus that Bjorn Lundberg [ph] organized, cosponsored by The Economist, and we'll be having an event on that on October 12. Scott Wallston [ph] will be doing an event out in California to which all of you are invited, but I don't think we are providing free travel, on the future of telecom, and we are doing that jointly with Stanford, which is why we are going to be holding it in beautiful downtown Palo Alto. In December, we're going to have an event on information markets and how they can impact public policy.  If you remember those terrible terrorist markets that were proposed by some pointy-headed academic and shot down by a non-pointy-headed politician, you may be interested in attending that event. For the last few months, partly because of you folks in the audience, we have been lucky enough to have had a very high level of downloads of our publications at the Joint Center. We are now enjoying over 50,000 downloads a month, probably because our downloads are free, so Scott Wallston and I are looking into the possibility of taking a small commission on each download. It may also be because Professors Sunstein and Viscusi have contributed quite liberally to our website. As most of you know, the Joint Center is no stranger to controversy, and that is why you shouldn't be surprised by today's title, "Is Everyone's Life Worth the Same?  Dilemmas for Regulators." Of course, it is not just dilemmas for regulators.  It is also dilemmas for almost real people like you and me. This controversy arose early during the Bush Administration--and I see several of you in the audience who have written about it--related to a regulation regarding particulate matter, where the issue was how should we count "old people"--and I use that phrase in quotation marks since I am over 50 now--versus young folks, and in particular, should we discount the value of a statistical life for older folks. The decision of the administration or the analyst who did the analysis was that old folks were worth less than young folks to the tune of 37.8 percent.  That is to say, there was a "senior death discount" as the press called it, and then the administrator quickly followed suit and said there shall be no "senior death discount," so I guess it disappeared.  It's kind of like a campaign slogan that you might hear on "Jay Leno" that he might have fun with.  Our session today is going to address some of the thorny valuation issues, but I wanted to spend a moment with you just trying to set the playing field, and then Professors Viscusi and Sunstein can correct me. The way that economists look at this issue is based on what folks are willing to pay for a small reduction in risk, and that is really important.  So, suppose a 40-year-old were willing to pay more than a 70-year-old person for a safety device that would decrease their fatality risk by some small amount.  Then, economists at least would say that the 40-year-old has a higher willingness to pay and therefore the benefits of making the purchase for the 40-year-old might be greater than the person who is 70. So we look at what people are willing to pay for risk reductions, and then we put that into the benefits side of the equation. There are a few wrinkles, of course, and I am sure we'll get to many more of these.  First, not everyone would agree that willingness to pay is a reasonable measure for counting benefits, but economists are fairly adamant in at least saying that it should be part of the picture. Second, even if willingness to pay were the right measure, there may be problems with actually measuring it in practice. Third, there are probably lots of alternatives that brilliant law professors and economists might think of, and I assume we'll hear about some of those today. And finally, the not inconsequential matter of how should we value, or statistically value, the lives of younger folks.  In particular, I am thinking of children. I want to bring up one very controversial concept before we get the ball rolling which I have alluded to but I want to define in a little bit more detail now, and that is the idea of the value of a statistical life. Let me start with what it is not, and this is based on the work of Gibb [ph].  It is not placing a value on a particular person's life.  It is, however, as I said earlier, integrally related to what an individual would be willing to pay for small reductions in risk.  If we keep that in mind, I think our discussion in the Q & A, which I know both of our presenters want to encourage and engage in, we'll be more informative. So let me present an example.  Suppose a person could reduce his mortality risk--and this will be on the test--suppose a person could reduce his mortality risk by one in 10,000, and they are willing to pay $500, say, for an airbag.  What would we compute as their value of statistical life? Well, we would simply take $500 in the numerator, divide it by one over 10,000, and get some large number which, if I did my arithmetic right, turns out to be $5 million for the value of a statistical life. That's all the concept is--no more, no less.  There has been a lot of ink spilled in academic journals over it, but for purposes of today's discussion, unless my distinguished colleagues want to redefine it, that is kind of where I want to leave it for the purpose of our presentations. Let me briefly introduce our two speakers, who require no introduction, and I would prefer to yield the floor to them, so I'll try to keep my introductions short. Cass Sunstein, who is immediately to my right, is a professor at both the University of Chicago Law School and also in the Department of Political Science there.  He is the author of many, many books.  One of my favorites is "Risk and Reason," which provides I think an extraordinarily insightful review of what we know and don't know about cost-benefit analysis and how it is actually used in practice. He is also authoring a forthcoming book on what I think you will find to a very interesting subject, entitled, "Laws of Fear:  Beyond the Precautionary Principle." Professor Sunstein has frequently worked with Members of Congress and the executive branch on issues of risk regulation, and we will be hearing from him shortly. Kip Viscusi, to my far right, is a professor of law and economics at Harvard Law School.  He is the author of a virtually uncountable--and I mean that sincerely--number of articles on the value of statistical life.  His work on this subject serves as the bible for regulators in the United States as well as worldwide. He has published widely in economic journals.  He started his own journal--I guess he needed another place to publish or there was a paucity of places to put good research on risk and uncertainty--and it is now the leading economics journal in that field. He has also written numerous books on risk perception, on litigation and regulation, and on managing Superfund, to mention a few. I would be remiss if I did not mention Professor Sunstein and Professor Viscusi's greatest claim to fame and material well-being, and that is that they both serve as academic advisors to the AEI-Brookings Joint Center, and for that we are very grateful. We flipped a coin to see who would be the lead-off speaker.  Professor Viscusi called the coin flip "heads," and the coin actually landed on its edge, but we are going to give him the benefit of the doubt. Professor Viscusi. PROFESSOR VISCUSI:  Thanks, Bob. I recently did a paper on the value of life and how it varies with age, and Alan Krupnik, who is out here, pointed out that he and everybody on this panel would be in the "near elderly" category.  So we'll have to change the name of this, what we call people. I will only give a brief overview, and hopefully, you'll ask some questions, so I'll talk--I was given 15 minutes or so. I'm going to retrace some of the ground that Bob Hahn already did.  So, how should we value risk to life?  Originally, government agencies used the present value of lost earnings, so the U.S. Department of Transportation was one of the leaders.  After people are killed in car crashes, we give them the present value of lost earnings minus consumption, so that was the approach that all government agencies used until the 1980s. There is the advantage that this number is easy to calculate.  It is used in court cases.  But it doesn't really make a lot of sense.  The benefit principle that should apply to all policies is a society's willingness to pay for the benefit.  And the benefit in the case of these risk policies is a reduced probability of death for people, so what we really want is society's willingness to pay for the reduced probability of death, which typically will be much greater than the amount you would pay off after they are dead in an automobile crash or some other accident. So there has been a cottage industry that has developed analyzing what the value of statistical life is.  Some people ask questions of others, like Bob just did here--how much would you be willing to pay for a small risk reduction--but for the most part, people have relied on market evidence pertaining to the value of life. So in the case of the labor market, how much extra do workers get paid for jobs that pose extra risk; for cars, how much extra do people pay for cars that are safer; for houses, what kind of discount do you get if you live near a hazardous waste site that may give you cancer, or if it is in a polluted area.  Seatbelt use--people have imputed value of life estimates from that. So there is a whole series of contexts that people have analyzed using real decisions.  And there have also been a number of survey studies eliciting value of life, willingness to pay estimates. I have done a couple of surveys of this literature--one was in 1993, and most recently, there was one with Joseph Aldie [ph] in which we surveyed all the surveys of the value of statistical life throughout the world, including a dozen other countries, and the consensus estimate for the United States is about $7 million. So my old number updated for inflation was over $6 million, and I think that shows up in some regulatory analyses, but my best current point estimate is about  $7 million per statistical life, which is a fairly substantial number. However, it is not one number, and not every study comes up with $7 million.  One reason is that there are heterogeneous preferences.  So, some studies include workers who are only in high-risk jobs, so workers in high-risk jobs will tend to self-select into those jobs because they are more willing to bear risk, and they will tend to have lower values of their statistical life. I have done with Joni Hirsch [ph] work on smokers.  Smokers are more willing to accept dangerous jobs.  People who wear seatbelts are less willing to accept dangerous jobs.  So there are a lot of differences out there in terms of people's tradeoffs between risk and money, but the average out there is about $7 million. It also varies with income.  So we have estimated, based on all the studies out there, there is an income elasticity of the value of life of about .5 to .6.  So a 10 percent increase in people's income will lead to a 5 to 6 percent increase in the value of statistical life. So it does vary by income group. Another reason studies get different results--different samples and different methodologies. Valuing life for policy--it used to be present value, and I first started to try to get government agencies to use this when I was in the Carter Administration and I worked as Deputy Director of the Council on Wage and Price Stability.  We used to do regulatory oversight. Tom Leonard is here and was also there, working with us.  And we tried to convince OSHA then, why don't you use these willingness to pay, value of statistical life numbers.  Their attitude was that putting a dollar value on life is immoral, so they wouldn't do it. However, in 1982, they ran into a problem.  They had a hazard communication regulatory proposal.  They calculated the costs and benefits, sent it over to OMB.  When they calculated the benefits, life was too sacred to value, so a life was only worth the cost of death.  In their view, the cost of death was the present value of lost earnings and medical costs if you died on the job.  So it was a reasonable small number. OMB tinkered with some of the assumptions and said, well, you messed up on a few things, and if you did it right, the costs are greater than the benefits. I was asked to settle the dispute between the two agencies, and if OSHA had done it correctly and actually valued lives according to the willingness to pay numbers from the labor market, then, benefits exceeded the cost.  So the first use of the value of life by a government agency was one which actually enhanced regulations, made them look better because it cranked up benefits by a factor of 10.  So from that standpoint, again, it was greeted enthusiastically by regulatory agencies. So I think one reason for the popularity of it is not because it is economically a sound approach, but it gave them bigger numbers than before. Age effects--EPA did the illustrative analysis with respect to the Clear Skies Initiative.  They had photos of elderly people with signs on their heads--"Seniors on Sale, 37 Percent Off." Should young people be more highly valued?  What do we think about that?  Well, young people have a bigger quantity of life, so if you are 80 years old, your remaining life expectancy is a whole lot less than if you are 23 years old.  So the quantity of life does vary. You might well look at the value per discounted life-year, which is an approach that some people did.  The quality of life also varies as well, so as you get older, your quality of life declines--although I am happy to report that mine hasn't too much yet. But all this arguing about whether there should be differences in the value of life according to age, to some extent may be a moderation.  I think it is a side issue, that the real issue is trying to get regulatory agencies to pay attention to the value of statistical life numbers when making policy, not just when doing regulatory impact analysis. So I have a working paper with Bob Hahn and Randy Ludder [ph], an AEI--it is probably called a "report" or something--where we look at risk analysis and the cost per life saved for many regulations is way off the charts, nowhere near these $7 million numbers, and may have a counterproductive effect when you take into account the total risk implications of expenditures at that level. So before we get bogged down too much in arguing over whether the elderly's lives should be valued at $7 million or $5 million or $4 million, it would be great to get regulatory agencies on board with actually basing policies on the benefits and costs of these policies. Now we are back to the age stuff.  Here is a series of estimates.  I have a couple of recent papers where we tried to get a handle on age variations and the value of life, using different datasets.  This is using the CPS, so it is a large national sample to which we matched information on mortality risks that are specific to the industry and age group of the worker. As you can see, the result here is a pattern that is consistent with--there have been theoretical analyses of what it should look like, and it should be an inverted use, so the value of life, the statistical life, should rise over time, eventually reach a peak and then flatten out. In this case, people to age 22 have a value of statistical life of maybe $3 million.  If you are age 37, it is just below $7 million, and if you are over 62, it comes down to about $2.5 million. MR. HAHN:  Kip, can you explain why that pattern is expected or should be? PROFESSOR VISCUSI:  Well, it depends on the model.  If there are perfect capital markets and perfect borrowing and lending, you don't get this result.  So if there are capital market imperfections so you can't borrow and lend across time, this is the kind of thing that you get. Now, I did notice one thing that I though was disturbing about this, which is that my value of statistical life has already peaked.  So I say we have got to work on this a little more. Another thing I noticed is that I go out and buy cars with all kinds of safety benefits-- every safety gadget known to man, I'll go off and purchase it--whereas my children, who are age 24 and 21, will buy topless Jeep Wranglers and tool around in those.  And it struck me that maybe these results aren't fully reflective of things. One thing they don't reflect is changes in your wealth.  The CPS doesn't have information on wealth.  So in another paper I have done with Tom Kneesner [ph] and Jim Ziliac [ph], we added your consumption to the equation, various measures of consumption, and for people who worry about the statistics of this, we did nice instrumental variables, estimates that have nice properties. What happens is that we are able to show the value of statistical right peaks a little further over to the right and does not drop as much when you age.  So that it actually ends up being fairly flat.  So it does rise and then subsequently fall, but if you are age 62, your value of statistical life is higher than the average for the U.S. population.  I think in survey results, Alan Krupnik and his colleagues are getting things that are quite consistent with that. Anyway, we did these illustrations of what this means for the Clear Skies Initiative.  There are two sets of estimates that they did--long-term exposure and short-term exposure.  Using a constant value of life let's you do the long-term exposures--$11.6 billion worth of benefits for adults age 18 to 64; $36.6 billion worth of benefits for adults age 65 and older. So you can see why this was a big ticket item.  If you adjust the benefits to those age 65 and older, and the next thing, value of senior adjusted--this was the illustrative calculation done by EPA--the 37 percent discount knocks of a whopping $13 billion worth of benefits. So it was a big player because lots of the benefits were concentrated among the elderly.  However, in the final column, we use our consumption-adjusted value of statistical life estimates, and we find that using those, you actually slightly increase the total benefits, because if we use the value of life for those like workers who are 62 years old and call that the value of statistical life for adults 65 and older, then we get a consumption-adjusted value of statistical life of $37.1 billion, which is $14 billion more than the senior-adjusted EPA number and actually half a billion more than the starting value number. So making age adjustments in the value of statistical life doesn't necessarily mean you have to sell off the elderly at a cut rate. In this case, they don't lose at all. So one thing you would want to take into account if you are doing these studies is not just to simply look at the total number of years of life left, but also take into account that people's resources vary a lot over time, and when we get back to the basic principle, which is that what we care about is people's willingness to pay for a small reduction in risk, there is no reason that that can't go up even if you have fewer years of life to live.  So your value of life does not necessarily peak at birth. Should income levels matter?  It matters for the lost earnings approach.  Willingness to pay increases with income.  Providing policies that the poor do not value is not necessarily a good idea. Back in 1992 when I did my first survey for the FAA, I argued that in the case of airline safety, since airline passengers have higher income levels than society at-large, we should regulate it more stringently than highway safety, particularly since it is the passengers pay for these safety costs through higher ticket prices, and it is not like the government is handing out the money. That argument was not successful, but I still think it is a good idea to regulate airline safety more stringently. How could the courts use correctly?  I'll just close with this--that if you are analyzing product safety decisions by the firm, lots of firms have gotten into trouble when they have done analyses of engineering decisions in which they impute a value of life equal to the amount that you pay out in the typical court case for killing somebody. So Ford got in trouble with the Pinto, although it's unclear exactly whether the characterization of that in the press was accurate.  GM certainly got in trouble for the Edward Ivy memo with respect to the gas-fed fuel fires in trucks.  And the problem with their analysis is not that they did a risk analysis, the problem is that they didn't use the right number. So instead of using the amount you pay off in a court case, when you are doing a design change and analyzing the cost, you should look at the value of statistical life and prize the lives that are saved using those numbers. That's it, and I'll be happy to answer questions after Cass Sunstein is done. Professor Sunstein? PROFESSOR SUNSTEIN:  From the lawyer's point of view, there is a real oddity, which is that the United States has two parallel systems for the valuation of mortality.  The first is the common law system, which in almost all States accords zero value to the loss to the deceased person, the hedonic loss of life, that is valued at zero in almost every jurisdiction. There is a high degree of variability in wrongful death actions and in survivors actions, and the legal system doesn't just tolerate that degree of variability, it invites it, because it pays close attention to the wealth of the person who has died, it pays attention to the number of dependents who were involved in the lawsuit, and in many States, the suffering of the decedent before this decedent dies, matters, although the hedonic loss of life doesn't matter. So what we get in the common law system is a lot of noise, which no one likes, but we also get a high degree of intended variability so that there isn't a single value, there are disparate values, and that is what the system is after. By contrast, the regulatory system has at least an aspiration to uniformity although some variability across agencies, which seems inexplicable.  Some use some Viscusi studies, others use other Viscusi studies, so they get variability because they are out-of-date; they haven't kept up with his work. But there is a consensus now within the agencies that the number should be $5 or $6 million for value of statistical life, and that is a uniform number that doesn't vary at all. What I am going to suggest is that in principle, there ought to be much more individuation than the current Federal regulatory system demonstrates and that that individuation is required by the very theory that underlies current practice. So this is just a suggestion, that if we are basing value of a statistical life on willingness to pay studies, we ought in the second generation debate, when we are trying to get good numbers rather than just decent numbers or some numbers, in the second generation debate, we ought to be moving to a high degree of differentiation across both risks and persons. One way of getting at this is just to notice that every individually, really, in  this room or in America or in the world has not just one value of statistical life or one willingness to pay but a large set of willingnesses to pay depending on the particular mortality risk in question. So if we look at individual behavior on the part of Americans or people here, we'll see that the willingness to pay to avoid a risk, let's say, of death in a car accident or death in an airplane or death from a consumer product or death from AIDS, that these willngnesses to pay are variable across risks as well as variable across persons, both because there is heterogeneity in preferences and because there are differences in initial endowments of various sorts. So it is kind of comical to think that there is one value of statistical life that captures individual behavior across ranges of differences, and the two dimensions on which we ought to focus are risks and persons, and we'll see differences both across risks and across persons. One way to make this point is to say that if we had a super-technology that had hedometers [ph], let's say, or willingness to pay meters that could perfectly capture each person's willingness to pay for each risk, then we would be perfecting the theory that underlies existing practice.  That is, we would be building on it, but making it much less crude than the single number is.  And then, we would be providing people with regulatory protection that perfectly matches their willingness to pay for the particular risk that they face. Now, if that is right conceptually, then the real gap is not theoretical, it is an empirical gap.  We just don't know what the numbers are.  Kip has given some figures, which is the kind of state of the art on distinctions across risks and persons, and I'll say a little more about this, but the gap is an empirical one, not a conceptual one, at least according to the theory that the government is now using.  I raise some questions about the theory that the government is using, but I think it is basically sound. So let's be a little more specific about risks and persons.  Many of the studies on which agencies now rely are based on studies of workplace accidents.  Not all of them, but many of them try to figure out how much workers are paid or how much workers are willing to pay to alter a risk; how much are they paid to be subjected to somewhat higher risk levels.  And then you have the simple exercise of multiplication, which elicits what is called the value of statistical life but which is more precisely the value of statistical risks.  That's what we're actually talking about. Now, workplace risks are not all the same as one another.  People who are working as police officers face different risks than people who are working as secretaries, and secretaries face different risks from the risks faced by firefighters, and firefighters have different risks from the risks faced by people who work in mines.  And as some of the data that Kip has compiled and others demonstrates, we find actually a different value of a statistical life in different occupations.  The numbers are not identical, and it would be kind of a surprising coincidence if they turned out to be identical. There are three differences that it is obvious that we should expect, or intuition suggests that we would expect them.  The first is that cancer risk ought to elicit a higher willingness to pay than other sorts of risks.  A lot of psychological work suggests that the risks associated with cancer are especially dreaded, and people are willing to pay a premium to avoid them.  Unfortunately, the psychological work is mostly contingent valuation studies rather than actual market behavior. Nonetheless, it wouldn't be at all surprising and there is some reason to think that there is a cancer premium out there, and if there is, agencies ought to use it. Kip has suggested the possibility of a difference between willingness to pay for air travel risks and willingness to pay for, say,  highway safety risks.  That might be a product of wealth differences between people who travel on airplanes and people who use highways, but even if that is not involved, there is reason to suspect--we would have to find out if it is so--that people are especially concerned to make sure that their flying is safe.  So a statistical risk of one in 500,000 to avoid a risk of dying in an airplane will elicit a higher willingness to pay plausibly than an equivalent risk of dying in a car accident. If that is so, then the uniformity of numbers between the FAA, let's say, and NHTSA, there isn't that.  This is inexplicable variety.  But if there were uniformity, that wouldn't be justified, because the theory suggests a reason for disparity. The third example I will give, the simple, intuitive one, is that different sorts of deaths produce different sorts of alarm, so a death from AIDS produces a higher willingness to pay, some contingent valuation studies suggest, than a sudden, unanticipated death.  And you'd think a death from Alzheimer's disease would produce a higher willingness to pay than a death from, let's say, a heart attack.  At least, it is plausible to think that's so. If this is right, this is another risk where there should be variability because the particular risk in question elicits higher willingness to pay. Such data as we have does suggest a degree of variety across statistically equivalent risks, and maybe there are methodological reasons for that, but the theory suggests that there will be at least some variety, and in the second generation of going beyond the uniform numbers, agencies ought to be picking up on such variety as the studies demonstrate. It is more controversial to suggest that there should be variability across persons, not just across risks, but let's just go over five categories of differences that we would expect to find. The first has to do just with taste.  If you have a risk that is faced by people who are particularly averse to statistical mortality risks, then we would expect a higher willingness to pay than if the risk is faced by a different sort of population. So insofar as there is preference heterogeneity with respect to risks, and this maps onto different regulatory programs, then, if we have the theory that we are using for willingness to pay, we should expect different numbers.  That is the first. The second Kip has referred to, which is age differences, and study studies as we have suggest that it is just not the case that people who are 30 and people who are 50 and people who are 70 show the same willingness to pay to avoid a statistically equivalent risk.  If that is so, then, OMB was on the right track in suggesting that there is grounds for disparities across different demographic groups defined in terms of age. More sensitive is the area of areas of wealth, race and sex, but such studies as we have do suggest that there are demographic differences--with wealth, it seems easiest and obvious--with respect to willingness to pay, the different racial groups which show different willingness to pay, and that men and women might show different willingness to pay.  It might be that the racial and sex differences would be purely an artifact of wealth differences, maybe not, but it might be that women would be more concerned about some risks than men are, or less concerned about some risks than men are, and that would map onto different willingness to pay numbers, producing statistical variations in VSL, value of statistical life. All I am saying here is that the theory that underlies current practice calls for agencies to adapt to those differences if they are empirically demonstrated. The last example--I have mentioned taste, age, wealth, race and sex.  The fifth I think is the most important, and that is international differences.  The theory that underlies willingness to pay and value of statistical life suggests that the value of a statistical life ought to be highly variable across nations, partly because of different risk preferences but probably most of all because of differences in wealth. So in South Korea, the value of a statistical life, a 1990 study suggests, is $800,000.  In India, it is about $1.2 million.  Whereas in the United States, it is $6 or $7 million.  In  Hong Kong, it is $1.7 million. Some critics, especially in the context of valuations of global warming, have been alarmed or appalled by the suggestion that there should be international differences, but if the data supports that, that is what ought to be done.  It does not make sense to apply to South Korea the same value of a statistical life that is found in Canada or the United States or England. What I have done thus far is just to suggest that given the account that agencies are now operating under, it is almost comically crude, at least after we have been at it for a while, to use a single number for a value of statistical life and the data that supports the $7 million figure or $6 million figure also supports an increasing degree of heterogeneity as regulators get more sophisticated about it. Now let's talk a little bit about the normative questions, whether the theory is right or whether this more differentiated application of the theory is subject to moral objections.  The first and most expected objection would be that differences certainly across persons with respect to age may be wealth, may be more race and sex, almost certainly; differences across persons offend an equality principle.  And some people might say the same is true for differences across kinds of risk, but that is less offensive, I think, to intuitions about equality.  It is objectionable, it might be thought, to value let's say Hispanics at a lower amount than Caucasians, even if the willingness to pay number supports that distinction. I think we have to be careful with the equality objection, for two reasons.  The first is the willingness to pay notion encodes equality of a sort.  That is, it says that people should be provided with protection in accordance with the economic values that they themselves show.  So it embeds an equality principle in it, and the differences that we see across persons are consistent with that form of equality rather than a violation of it. Here is another way to put it:  Poor people don't buy Volvos.  Wealthy people buy Volvos.  Does that offend the equality principle that Volvos go to richer people and not to poorer people?  If not, it is because the satisfaction of people's preferences with respect to the tradeoff between automobile safety and other commodities, allowing people to make that tradeoff as they see fit given their income constraints, that is equality, that is not a violation of equality. That might be maybe too semantic or fancy a response, so let's try another one to the equality objection which would say that poor people, or people who have less resources than others, we don't do them any favor by forcing them to buy safety at levels that exceed their willingness to pay.  So it would not be a terrific benefit for people in China or India to tell their government to value statistical lives at the level that Africans and Canadians value statistical lives.  That would be a way of buying regulatory protection for people in China and India in a way that would defeat other goals that China and India have.  Is this point clear? It is analytically not different, I think, to use the U.S. value for China and India than just to double the U.S. value in the interest of helping American consumers.  Would it be great for America if we said instead of $7 million, it would be $14 million?  Not really, because we would be forcing people to buy double the protection against statistical risks of one in 500,000, one in a million, one in 200,000, then we would be forcing them to do the equivalent of buying Volvos.  Is that in their interest? This last suggestion which sees regulation analytically as a kind of forced exchange in which people are being asked to buy goods at price that maybe exceed their willingness to pay, which isn't ordinarily a great way of helping people, this suggestion suggests the importance, I think, of distinguishing between easy cases and hard cases for the use of the willingness to pay criterion. The easy cases, which I think if we are clear on this, everything gets off the ground at least, are those in which the beneficiaries of regulation are paying all or almost all of its cost. So think of cases--workers' compensation seems to take this form; drinking water regulation seems to take this form; there are other sorts of regulation that have the same features--ones in which people are getting regulating protection, and they are paying for it 100 pennies on the dollar.  So for every dollar of protection they get, they are paying all or almost all of the cost. If that is the situation, then the case for use of willingness to pay criterion is very straightforward, isn't it--unless there is some problem of lack of information or bounded rationality or something on the part of consumers or workers. Why ought government to force people to pay an amount for regulatory protection that exceeds their willingness to pay?  And if regulation has this form of the forced exchange, workers' compensation, arsenic regulation and so forth, then the theory is very straightforward, and we ought to have as much individuation as we can feasibly get given the empirical findings that we trust. A harder case is one in which the people who are getting regulatory benefits pay only a fraction of  its costs, and I think many environmentalists and occupational safety and regulation fans think this is the dominant situation.  Let's talk a little bit about this one. This is one in which let's suppose the--okay, I'm almost done--by edict--let's suppose the beneficiaries of regulation are paying just a fraction of its cost, and they are getting something which costs more than their willingness to pay but from which they are net winners. So let's suppose that poor people in Chicago, let's say,k are getting clean air; its provision costs, let's suppose, $100 per unit.  They are willing to pay only $80, but they only have to pay $20.  Is that clear? This is a situation in which the beneficiaries of regulation are net winners even though the regulation isn't efficient, because the willingness to pay of the beneficiaries is lower than its social cost. I think this situation, the hard case, is actually analytically trickier.  It may be that the distributional gain to the people at the bottom justifies the inefficient regulation--although this isn't the best way of helping people at the bottom, it is a way--and it may be--and I can elaborate this more later if you like--it may be that if we are concerned about welfare rather than efficiency, this regulation might actually promote welfare goals. So this is just a suggestion that the easy case is the simplest one; there are harder cases--but even in the harder cases, a nonvariable uniform willingness to pay need not provide much help because if you are valuing people at an amount that exceeds their willingness to pay, we don't know that we are getting any distributional gain.  It may be that poor people are the victims, and wealthy people are the principal beneficiaries--think, for example, maybe of airline safety regulation which has a high willingness to pay.  There may be special cases we can identify where we will embed distributional considerations into our ultimate decision, but a good way to do that is not to insist $7 million, $6 million, $5 million. In other words, a uniform willingness to pay value of a statistical life number is a crude redistributive strategy. So the upshot is regulators, according to the very theory they are using, ought to have a much higher degree of individuation than they now do.  In this way, at least they can build on the common law and not, incidentally, the September 11 Compensation fund, which makes distinctions, doesn't treat everybody the same.  And if we want to subsidize people who need help, let's do that and not have a regulatory system that has a value of statistical lives that is artificially inflated.  That is not a subsidy. MR. HAHN:  Great.  Thanks, Professor Viscusi. Do you want to say anything, or do you want to react to Kip's presentation at all, Cass? PROFESSOR SUNSTEIN:  Well, I thought that we kind of marched hand-in-hand, and what we need is much more empirical work about individual-- [TAPE 1, SIDE B] PROFESSOR SUNSTEIN [continuing]:  --and it seems that agencies are finally starting to do that and so eager for them to do this more seriously that this second-generation debate might seem kind of exotic.  And I see that, but it's good for the regulators to get it right, and if we can build up less crude mechanisms for analysis, then, sooner or later, they will probably be adopted. MR. HAHN:  We want to hear from you, and I'm going to ask Sasha going around--and I think we're taping this, so if you could wait for the microphone, raise your hand and identify yourself and direct your question to either Professor Viscusi or Professor Sunstein. QUESTION:  Bonnie Wachtel [ph]. I think this is for Professor Viscusi.  I like the theoretical construct that you have set up, and I am curious about some of the information questions. In other words, you look at a person--how much additional pay do they require to become a fireman from whatever else they are doing--but I wonder if the person, when they are making that decision, is actually handed a piece of paper saying, "Now, statistically, your rate of mortality is going to be this much higher if you look to the fireman's pay." It seems to me the decisions that we make in our ordinary lives are really sort of gut reaction decisions, and I have even read that people are notoriously poor judges of gauging statistical risk.  They think it's worth money to not have alar on an apple, or something, when there is about a one in one billion change that that would hurt you from a health and safety perspective. PROFESSOR VISCUSI:  Sure.  In the case of alar, highly-publicized, small risk, people may develop inaccurate perceptions. Job risk, people can observe lots of things about the job, and nobody hands them sheets of paper telling them the risk. I have done studies on chemical workers at four different chemical plants, and their assessed risks were right on the money when we actually interview them and elicit their risk assessments on a probability scale. What is interesting is that the market studies generate numbers that are very, very similar to the ones where you actually tell people what the probability is in a survey context and elicit their willingness to pay.  You also get similar results across different kinds of risk.  So I have gotten results for cars, for houses, seatbelts and for job risks that come up with very similar numbers. So I think when people are asked whether they are facing real risks, they have a general sense that, you know, working construction is a lot more dangerous than being a secretary, for example.  And based on my experience--I worked two summers on the assembly line at General Electric--and if there was a fatality in Building 1, you heard about it.  So information about injuries did get around, and the company posted the number of days without a lost workday statistic. QUESTION:  Two factors that you didn't mention in variability that I think are interesting.  One that has been cited a lot is voluntariness of risk, and another which is closely related which I have heard discussed less is property rights, whether a risk is being imposed on you that you think violates your property right versus a willingness to pay to avoid a risk where maybe the property right lies with the risk or where you don't feel you have a property right to have that risk. Can you comment on those two factors? PROFESSOR VISCUSI:  Do you want to give an example of the particular property right context you might be thinking of? QUESTION:  Well, with the example of the car, the Volvo, I don't think most people would say they have a property right to safe highways that is somehow being violated, so they are purchasing something rather than--it's not something they feel an entitlement to.  Whereas, let's say an industrial pollutant that poses a cancer risk on surrounding communities, people might think that that violates a property right of the communities. PROFESSOR VISCUSI:  Sure.  In the case of voluntary versus involuntary, one reason you get smaller estimates of the value of statistical life for voluntary risks is that people who are more willing to bear risk tend to gravitate into those situations.  You get different people matched up to those risks. So if you are looking at government regulation where you are protecting people against involuntary risk, you'd think that the average value of statistical life for that pool of protected people would be greater than if you are protecting people who have knowingly chosen to incur large risks, because there is substantial heterogeneity out there. If you look at workers in the construction industry, their value of statistical life is substantially below that of the working population at large, and their risks are 10 times greater as well. As for the property right itself, I have done work on--you mentioned hazardous waste sites--we have estimated that--Ted Gayer [ph], who is now two days a week at AEI and the rest of the time at Georgetown--the value of statistical cancer cases associated with being exposed to Superfund sites, and for that, we found numbers that, according to your analysis, should be bigger than $7 million, but they aren't; they are basically in the same ball park. So people express a high willingness to have the government spend money to clean up Superfund sites--so if it is other people's money, there is an unlimited willingness to pay for this--but in terms of the housing price hit that people actually do take because of this, the value of statistical life number comes out about the same. PROFESSOR SUNSTEIN:  A couple things.  On the voluntary/involuntary, there is a lot of data on this, and it supports the distinction.  I think it is a puzzling distinction.  Okay.  An example of a voluntarily run risk, I guess, would be driving in a car that's less safe than other cars.  An example of an involuntarily run risk might be flying or living in Los Angeles.  I guess these are different, but what does this mean, this distinction between voluntary and involuntarily run risk?  You do have a choice whether to live in Los Angeles, and you have a choice whether to fly or not.  So I'm not clear that the sharp dichotomy between the two is really what is driving these answers.  It might have something to do with risk preference, or it might have something to do with the cost of risk avoidance.  We characterize things as involuntary where it is very costly to avoid the risk, and voluntary where it is cheap to, maybe, which suggests it is not a dichotomy. So this distinction which a lot of people are excited about, I wonder whether it is as sharp as suggested. I was wondering if by your question you meant the endowment effect--and Kip has done some work on this--where, if you ask people how much you are willing to pay to avoid a risk of one in 100,000 from apples, you get a lower number than if you ask how much would you have to be paid to be subject to a one in 100,000 risk from apples.  You've got a distinction.  And that seems to be a pretty robust findings, and I think it is an intriguing finding, that people are willing to pay less to avoid a new risk than they would demand to be subject to a risk. This may be a form of quasi-rationality.  There may be some problem with this disparity.  What is great I think about the market studies is they avoid this problem, because when works get let's say $60 for a risk of one in 100,000, it's hard to get your mind around is this willingness to pay or willingness to accept.  It doesn't really matter. So I think the distinction is relevant for some contingent valuation studies; it is not clear exactly what to make of it.  OIRA says use WTP usually, not willingness to accept.  And I think there are problems with the willingness to accept number, signaled maybe by the fact that if you ask many people, a certain percentage of people say "There is no amount that is large enough to allow me to be subject to an additional one in 100,000 risk of cancer--are you kidding?  You can't pay me anything for that."  And that suggests there's something funny going on in some of those answers.  But you are right. MR. HAHN:  Up here at this table--Jordan.  If you could identify yourself before you ask your question, I'd appreciate it--and the institution.  Professor Viscusi wants to be able to track you down. QUESTION:  Sally Katzen. I don't have an institution as such. MR. HAHN:  You are an institution. QUESTION:  I wanted to follow up on what Cass was just saying--and Bob, you may say this is out of order, since your ground rules originally said "Economists say willingness to pay, period.  We're going to accept that, and that's going to be the basis." But willingness to pay is clearly constrained by your own resources.  What somebody is willing to pay who is getting minimum wage is different from what Bill Gates is willing to pay.  And to discard the willingness to accept because you get these incredible numbers, well, those are not constrained by resources.  Those are really what people feel. I was intrigued with you saying, "There are these problems, and we'll get rid of them."  Why hasn't there been more attention to trying to refine the willingness to accept, since it doesn't have the resource constraint? PROFESSOR SUNSTEIN:  Well, you might be nervous about willingness to pay numbers generally, because willingness to pay is an artifact of ability to pay, and people without a lot of money aren't going to have much willingness to pay, and aren't their lives to count for less, it might be asked rhetorically. I think it's not a rhetorical question, because if poor people are wiling to pay little to be protected against a statistical risk, you don't do them any favors by forcing them to devote their little resources to elimination of that risk when they have other things to do with that. That's why I'm saying it's very important to started with the forced exchange case, and in the forced exchange case, if you are poor, and you are told you're going to have to pay $60 to eliminate a one in one million risk, that's not good; that doesn't do poor people favors.  If we are talking about subsidy programs in which the taxpayers, let's say, are paying for the elimination of risk, that's another situation.  But regulation isn't typically a subsidy program. Isn't this clear?  I think what-- QUESTION:  Yes, that sounds intriguing, but I am wondering why we can't take into account more than normal constraints on resources. PROFESSOR VISCUSI:  Let me just add that the labor market numbers are willingness to accept numbers, not willingness to pay numbers.  So these are the compensation for risk. I get the goofy numbers in terms of the spread between willingness to pay versus willingness to accept in survey studies where there is simply an alarmist response.  We did this back in the eighties.  We gave people different hazardous chemical products and asked how, "How much would you be willing to pay to reduce the risk of let's say hand burn by one chance in 10,000." "I'd pay an extra 20 cents a bottle." "What if we reformulated the product so that it increased your risk of a hand burn by one chance in 10,000?"  "Oh, I'm not going to buy that at any price." So that part of it may be that just from running a study, it's hard to get sensible answers to willingness to accept in a survey context. PROFESSOR SUNSTEIN:  I think you think willingness to adapt elicits a more accurate number from poor people, but I wonder.  If you are a poor person, you ought to be willing to accept less, wouldn't you, to be subject to a risk, because you need the money?  And I don't know that there is any evidence that poor people show higher willingness to accept numbers than rich people.  That would be surprising, wouldn't it? So I am with you that distributional considerations seem to me an important part of regulatory policy in cases in which the regulation is helping people who are suffering.  But the use of a high value of statistical life for poor people may be a really lousy way of helping them distributionally.  You'd need to know the incidence of costs and benefits. Another way to put it is, okay, suppose there is a regulation protecting farm workers how are poor, and the value of a statistical life used for them is $9 million, let's suppose, even though their actual value of a statistical life is $2 million. Is this great for them?  Not if the money comes out of their pockets. QUESTION (MS. KATZEN):  Which is why I thought it was useful to distinguish the easy cases where the beneficiary pays most or all from the other.  But I have taken enough time. Thank you. QUESTION:  Yes, Al Krupnik [ph] from  Resources for the Future. I guess this is to both of you guys.  It seems to me there is another risk characteristic that is interesting and that I struggle with in my work as well, which is public goods versus private goods. So in my work, we are asking people their willingness to pay in surveys.  We ask 70-year-olds and up their willingness to pay to reduce their risk, we ask people younger than 70 their willingness to pay, and we do sometimes find lower values for the older group--the true, perhaps, elderly as opposed to the near-elderly-- PROFESSOR SUNSTEIN:  That's a very [inaudible] concept. QUESTION (MR. KRUPNIK):  --we do find those differences.  But then, if you ask, I think, most people--and so that would suggest you should do a senior discount--but if you ask most people if they would put a lower value on senior lives, programs that would save seniors' lives versus other lives, they may not say that.  And in fact the seniors may not say that.  I don't know.  There are public preferences that may in some cases trump private preferences. Now, among economists, we kind of have a way out of this issue by talking about paternalistic altruism and nonpaternalistic altruism, and that is one way to go.  But assuming this is the right type--or the wrong type--of altruism, what do you think about the issue of preferences individuals have for social benefits, benefits to others, versus, of course, in the private case that Kip is working on, that I work on, about their preferences for risk reductions to themselves? PROFESSOR VISCUSI:  This is the framing issue, so if you've run the survey, give people a choice between two policies--we can prevent 10,000 deaths among people with six months of life expectancy--they have advanced respiratory disease--or 10,000 deaths among third-graders.  Which would you pick?  Would they still say it's a tossup? PROFESSOR SUNSTEIN:  I wouldn't.  I have actually done that study, and you're right. MR.  :  Well, they would say the third-graders. QUESTION (MR. KRUPNIK):  Yes, but the studies may not--I mean, the studies for private preferences--well, you can't ask third-graders--but you could ask young people their willingness to pay, which would probably be pretty low, because young people think they will live forever. You  ask older people, and it will be probably higher than that.  Yet most people in society would say, well, let's save the younger people's lives. But yet if you use a strict private calculus, you wouldn't get there. PROFESSOR VISCUSI:  If it turns out that way, then I'd go with the private calculus, not this public opinion poll stuff. QUESTION (MR. KRUPNIK):  But it's not a public opinion poll.  It's preferences.  I mean, I have wrestled with this.  This isn't something I know the answer to, and I'm curious, Cass, how you would put that into your framework. PROFESSOR SUNSTEIN:  It's a great question.  The easy one I think would be suppose we aggregate preferences for race discrimination and willingness to pay, and it turns out there is a certain level of race discrimination that people are willing to pay for, and let's suppose it is clearly race discrimination of the sort that  is illegal. I think that has no moral standing because people's willingness to pay to discriminate against people is illicit and off-limits.  So too with--there is a set of issues where the willingness to pay analysis just isn't the right way to think about it because there is a moral dimension that it misses which has to do with illicit preferences, let's say. In this case, I'm not sure whether that's so.  When people say, let's suppose, that third-graders get a fantastic level of protection, we want to know what they are thinking when they think that.  Are they thinking that the resources that are protecting them are coming from the sky, or are the resources that are protecting them coming out of things that would otherwise be helping those children in some other way? So we want to know if there is some moral judgment that you are eliciting that is adequately informed and alert to the array of tradeoffs, then, that might trump the market judgment, but if not, no. I do think that the valuation of children's lives is extremely difficult because it is low, and I think this is more like Sally's case where, typically, a child's welfare doesn't go down dollar-for-dollar if the government provides regulatory protection of the child, although sometimes maybe that does happen.  So the children's one needs a lot more work. MR. HAHN:  Over here in the front. QUESTION:  Pat Kasano [ph], GE, and I'm not an economist, and it will probably show in my question. This seems very complicated to me, and I am wondering if there is not a simpler way of getting at it, and maybe it has already been done, which is the whole willingness to pay thing to me seems to be an attempt to try to predict what people will do.  And I'm wondering if there is a retrospective way to do this. Could you look at things that are relatively noncontroversial, for example, use of seatbelts and what we pay to install seatbelts in cars and what the American public generally speaking is willingness to pay to avoid that risk, and do a series of studies using things like that?  Maybe for children, you'd use immunizations, which are somewhat controversial, but most people generally speaking are willing to pay whatever it takes or have their child immunized at the expense of the government if they need to in order to protect against risk of various diseases. Could you take studies like that and use them to calculate what the value of a life is? PROFESSOR VISCUSI:  Well, I've done it with seatbelts using not the purchase cost of the seatbelt but time and disutility of buckling.  Glen Lindquist [ph] is the one who started the seatbelt stuff. But for children, some people have looked at bike helmets, like whether you buy bike helmets for children and how much that reduces their risk, and what's the cost of a bike helmet, and then imputing from that the value you place on children's lives.  But I agree with Cass, this is the toughest one to get a handle on, because it's hard for them to express their preferences.  How much would you be willing to pay if you knew your lifetime earnings profile, which is basically the type of answer we want to get. QUESTION:  Thank you.  I am Fran Smith with Consumer Alert. When we are discussing willingness to pay in the context of risk reduction for an individual, it seems to me that the person and the regulators are not looking at those tradeoffs that I think both of you pointed to. Wouldn't it be better for regulators, rather than looking atone risk in a vacuum and the reduction of that risk, to look at the risk versus risk--in addressing a risk, a regulator is trying to reduce that risk, but that action may in fact produce other higher risks.  And of course, we have examples of this--CAFE standards, delay in approving drugs, the DDT example, et cetera, et cetera. So by regulators looking at one risk in a vacuum, that is a bias that I think turns out to bias the whole regulatory action. I know both of you have done work in that area and on the precautionary principle and such.  Could you address that? PROFESSOR SUNSTEIN:  Yes.  It's an important point, and I think it is a less controversial point than what we're doing about. The precautionary principle which is so popular in Europe and in some places in the United States says in the face of uncertainty, regulate; that possible risks should be controlled by government. The problem with that is not that it is too regulatory or insensitive to cost--at least, that's not the most serious problem.  The most serious problem is it is incoherent, because any step you take to reduce risks will create speculate risks of another sort.  Try in the next 24 hours to reduce all risks simultaneously in your own behavior.  You might find yourself paralyzed and just standing there, but then you'd better start moving because there is no way to do it.  And the same is true for regulators. So definitely, it is very important for regulators to be alert to the possibility that the risk regulation they are engaged in will increase other risks, so we want to have net risk reduction.  But it may be that a policy intervention isn't increasing other risks--it's just imposing dollar costs. There is one difficult with that, because as both Bob and Kip have written, if you impose high economic cost,  you will probably create some mortality and morbidity effects as a result. So risk-risk analysis of the sort that you want calls for attention to cost at least indirectly, because a multi-million-dollar cost will create some statistical deaths. So for that reason, we want to attend to costs--it is part of risk-risk analysis--but I guess I'd say independently of that that if you are asking Americans to pay $500 to eliminate a risk of, say, one in 10,000, you should be nervous about that if people want to spend their money on other things and there isn't an absence of information or something, and you're not getting any distributive gain from that. QUESTION:  Howard Katz, Wagner Law School. When you talk about the difference in risk across people and across risks themselves, I am wondering if any of those preferences might be considered systematically irrational in a way that we might, from some objective basis, discount those revealed preferences. PROFESSOR SUNSTEIN:  Well, it's possible.  If you find that the cancer premium is extremely high, then you might think that the word "cancer" is producing a strong affective reaction that is very hard to explain. So some at least contingent valuation studies suggest that cancer death produces two or three times the willingness to pay of a sudden unanticipated death. Now, I don't believe there are any market studies that support that, and there are some that go the other way.  But suppose something like that were true.  At least, we might want to inquire whether the word "cancer" isn't calling up images of intense suffering that cannot possibly justify that degree of disparity. So yes, it is possible that some of the disparities can be impeached, but it would be very puzzling if uniformity across risks or persons--perfect or near-perfect uniformity--were itself required by rationality. PROFESSOR VISCUSI:  Let me chime in on this one that if you have different kinds of death, I can see why there would be substantial heterogeneity, because the morbidity consequences may differ.  If your choices across different domains are continuous--I've actually got a paper showing your marginal value of life should be the same across domains--the problem is lots of times, they are not continuous, they are discrete.  So you don't have a choice of--if you buy the bike helmet that you wear, that's sort of a lumpy decision.  So when we observe actual choices people make, even if it was the same nature of death, you may observe different values of statistical life just because our choices aren't smooth. But ideally, we should be looking at situations of trying to impute a value of life where people have a continuum of choices and not just one, single thing they can do. QUESTION:  Nancy Udell [ph], Common Good. I am curious whether you can talk a little bit, Cass, about your initial distinction between the common law system and the regulatory system and the different ways in which we value life in each.  I'm just curious about the application of this willingness to pay concept perhaps on the flip side in the tort system, keeping in mind your point that regulation is a forced exchange and that that is an ex post valuation as opposed to ex ante. PROFESSOR SUNSTEIN:  Yes.  I think the wrongful death awards are much too low.  A lot of them are high, and there is inexplicable variability in them.  But they ought to be at least around $7 million, shouldn't they, and they aren't.  Seven million is outlier high. So all States of the Union should move toward more uniformity, less noise, and higher awards in case of wrongful death. Now, having said that, that would be like what has happened in regulation in the first 20-odd years.  Then, we want them to be making distinctions so if it is a 30-year-old who gets killed, probably more than if it is a 94-year-old who gets killed. MR. HAHN:  We've got time for one or two more questions. QUESTION:  Dennis Zimmerman, CBO. I was curious--your age adjustment for wealth--should we expect that to vary by the strength of bequest motive in the sense that if that were to be expected, it is very important that the sample gives the estimates, includes the right share of things such as those with children and those without children.  Is bequest motive important in that valuation of wealth? PROFESSOR VISCUSI:  That's interesting.  We haven't looked at that, but you would think that if you want to leave your money behind,  you may want to spend less of it now. On the other hand, if you really care a lot about your children, you may want to hang around so that you are there to see them. So is it more valuable to leave them the extra thousand dollars you save by skipping the side curtain airbags, or should you spend the extra thousand dollars to protect yourself so you can be around to see your grandchildren a little longer? But yes, I think that's a neat idea.  I actually expected a smoking question from you given your past work.  But yes, it is something that would be great to look at, and we haven't done it yet. MR. HAHN:  One last question. QUESTION:  Thank you.  I'm Jane Leggett. I think my question is more directed toward Cass.  It seems to me that there are really potent redistributional issues in all of the questions you're asking about willingness to pay or willingness to accept or property rights, and that when you say--I'll give an example in what you said--when you gave the example are we really doing a benefit to farm workers to require them to pay more than they are willing to pay because it comes out of their pockets.  I was surprised at your positing it that way, because I'm thinking, well, consumers of those farm products are in fact the beneficiaries of the risks that the farm worker bears, and I was expecting you to say is it worth that some farm jobs will be lost as a result of that, as opposed that if it comes--and it crystallized to me that embedded in these, whether you are paying to reduce the risk to yourself or someone in your community or someone that you are related to or someone in another country would have a really strong bearing on your willingness to pay or your willingness to accept, or if a risk is imposed on you--let's say air pollution--and I could move anyplace, but I can't necessarily find a job where there is no air pollution--what would my willingness to accept be to bear the air pollution if it is coming from someone else versus who would pay for that. And I'm really confused, or I guess I'm concerned that we are embedding a lot of normative questions that involve important redistributional issues but not separating those out from what the right measure is and whether we are getting good measurements. PROFESSOR SUNSTEIN:  Okay.  Let me tell you what I feel fairly confident of and what I don't. I feel fairly confident that on the theory that the government uses now, a higher  degree of variability is required in principle.  I feel fairly confident that in cases involving forced exchanges, it's very hard to have an objection not only to the willingness to pay criterion as such but a high degree of individuation. Harder cases arise where the distributional incidence of a regulation might argue in favor of it even though the willingness to pay numbers suggest that people are getting more than they are willing to pay for.  So we may have cases in which the cost-benefit analysis suggests it's not justified in which let's suppose consumers are paying more for a good, but farm workers are getting safer.  And let's suppose the class of consumers are pretty wealthy people, and the farm workers are pretty poor.  It may be that on distributional grounds, that's worth it anyhow.  It is not--this is a somewhat technical point--but it's not the most efficient way to redistribute resources.  We would rather give the farm workers money than do this.  But as far as I am concerned, pretty good inefficient redistribution is better than none, and if this does it, that's nice. It also might be--willingness to pay I think has standing insofar as it is a proxy for welfare.  So if you are willing to pay $10 for something, that's a clue about how much welfare you will get from it.  But in these complicated distributional cases, it may be that the farm workers are getting a ton of welfare from the safety protection; the willingness to pay number is low just because they don't have much money.  So they might be getting a great welfare gain and consumers, though they are paying more, they might not be suffering so much of a welfare loss. That's why I am confused about cases of this kind.  Now, if we had a perfectly efficiency regulatory system plus an optimal income tax, then we wouldn't have to worry about all this, because we'd do all the maximizing through regulation, and then we'd handle the distributional problem through the income tax.  But we don't in my view have an optimally redistributive income tax. Okay.  But I think some skeptics about willingness to pay or individuated willingness to pay take the points just made--they take them too far.  They think this means that a high willingness to pay number is generally beneficial to poor people.  It isn't.  It all depends on the distributive incidence. So as you said, if the result of the high willingness to pay number for the farm worker protection program is to deprive farm workers of their jobs, then this is not only inefficient, it is distributionally perverse.  And now the informational demand on government to figure out what to do becomes awful, because they do not only the complicated cost-benefit analysis, and now I am asking them to individuate, and now you are asking them, I think correctly in principle, to take account of the distributional concerns. So for me, the upshot--you have been very patient with this complicated answer--the upshot is pretty straightforward, which is first generation do willingness to pay--that's great; second generation, start making some less controversial distinctions across risks if there is support for them, maybe make some less controversial distinctions across persons if there is support for them, and there is evidence that they are. Be alert to the fact that if there is some distributional gain to the most disadvantaged people, then we are not going to get all theological about the cost-benefit analysis.  But you had better be confident you're going to get a distributional gain rather than a distributional loss. MR. HAHN:  I want to thank you for joining us and asking some very provocative questions, and I'd like to ask you to join me in thanking our speakers today. [Applause.]</body></html>