Association Between Income and Life Expectancy in the United States

Hello and welcome to this podcast. This is Howard Bachner, Editor-in-Chief of JAMA. I am here with Raj Chetty and Angus Deaton. Raj is from Stanford and has written a paper for us entitled "The Association Between Income and Life Expectancy in the United States." Angus Deaton is at Princeton and he has penned an accompanying editorial entitled "On Death and Money, History, Facts and Explanations." Welcome, Raj. Welcome, Angus. Raj, let's start with you. Can you just take a few minutes to outline the study and then the four or five principal findings? In this study, we examined the relationship between income and life expectancy in the US using newly available data from deidentified tax and social security records. So we used information on about 1.4 billion observations for the entire population in the US between the ages of 40 and 76 during the years 2001 to 2014. And we used that data to construct estimates of life expectancy by income group and also by area in the United States at the county and commuting zone level. Marketing zones are analogous to metro areas and provide a partition of the entire country. Our analysis yields four sets of findings. First consistent with prior work at the national level, we find that there are very large gaps in life expectancy between high income and low income individuals. For example, men in the top 1% of the income distribution live about 14 years longer than men in the bottom 1% of the income distribution, to give you a benchmark to get a sense of that magnitude, men in the bottom 1% of the income distribution have life expectancy comparable to the average in Sudan or Pakistan, while men in the top 1% in the United States have life expectancy that's higher than average life expectancy in any country in the world. So there are vast gaps in life expectancy within America based on income. We also find similar gaps in life expectancy for women, however, an interesting finding is that the gradient, the difference between the rich and the poor, is smaller for women than for men. It's about 10 years. And in particular, we find a narrowing of the gender gap in life expectancy at high levels of income. Poor women live much longer than poor men. Rich women live only slightly longer than rich men. So those are some estimates at the national level. What we do next is then look at variation across areas within the United States. And one of the key new findings of the paper is that while most prior work on income disparities or socioeconomic disparities in health outcomes and life expectancy has focused on the national level, we find that the story really varies greatly across areas within America. So in particular, there are some places, for instance, New York and San Francisco where the poor have life expectancies above 80 years. So men in the bottom quartile live around 82 years in those cities. In contrast, in places like Gary Indiana or Las Vegas, Nevada, poor men's life expectancies are something like five years shorter around 77 years. So there are quite substantial differences in the level of life expectancy, especially for the poor across areas. For the rich, interestingly, it turns out to matter much less where you live. You find much smaller differences across areas. So in the next part of our analysis, we turn to trends in life expectancy over time. So the result I've been summarizing focused on the average levels of differences between the poor and the rich during the 2000s. Turns out that there are also sharp differences in the change in life expectancy over time. We find that overall, when you look at the country as a whole, the gains in life expectancy have been much larger for the rich than the poor. On average, the rich have gained people in the top income quartile have gained something like three years of life expectancy over the 2000s, whereas people in the bottom 5 percent of the income distribution have experienced essentially no gain at all on average. But once again, it turns out that this national story has a great deal of local variation to it. There are some places like Birmingham, Alabama, and Cincinnati, Ohio, where the poor gained almost as much in life expectancy as the rich. In contrast, there are other places like Tampa, Florida, and Knoxville, Tennessee, where the poor actually had declining life expectancy over the 2000s. And so that pattern we see of the poor staying roughly steady on average in the US as a whole masks the fact that there are actually some places where life expectancy for the poor is improving quite a bit. And there are others where it's actually going in the opposite direction. The last set of results turns to why we see this geographic variation, what's driving all these gaps in life expectancy across income groups, across areas. We don't have definitive answers here, but what we are able to do is identify some of the factors that are associated with having particularly high levels of life expectancy for the poor. And what we find is that areas with better health behaviors, so the strongest correlations we find are with measures of health behaviors. Places with lower rates of smoking among the poor, lower rates of obesity and higher rates of exercise have significantly higher levels of life expectancy. In contrast with the very strong correlations with health behaviors, we find much weaker correlations with measures of health care access. So if you look at the fraction of the population that has health insurance or the number of primary care physicians or measures of the quality of care in an area, we generally find no significant association between those measures and levels of life expectancy for the poor. We also find weak correlations with a number of other factors that have been discussed in prior work, including work by Angus Deaton. So we find very little support for the theory that differences in inequality across areas and income inequality are associated with differences in life expectancy, which is consistent with some prior work that Angus has done as well. We also find small associations with environmental factors. So is it that the poor, for instance, are exposed to more pollution than the rich or they live in areas with less access to good food, things like that? We find weak associations with proxies for environmental differences. We also find little association with labor market conditions, for instance, unemployment rates and measures of the health of the economy. So we end up concluding that individuals with low income tend to live longer in places where they have more helpful behaviors, which tend to be cities like New York and San Francisco, which are places that are affluent cities with a highly educated population with lots of immigrants. And the key question going forward is why exactly we see better health behaviors among low income individuals living in such affluent cities, and we believe that to future work? Thank you, Raj. Angus, much of your career is focused on these issues. The relationship between income and life expectancy is not new. Can you just talk about the literature on that issue and then reflect a bit on Raj's paper? I'd be delighted to, thanks very much. There's been hundreds and hundreds of studies going back a very, very long time, probably most notably at the beginning of the Industrial Revolution in the 19th century. There's B11A in Paris and Engels in Manchester and for a child in Germany. And all of those authors and many, many others looked usually across areas or across groups of one sort or another and found correlations between death rates and what in the literature is often called socioeconomic status. One of the problems with a lot of that literature is that socioeconomic status means different things in different contacts, but in many of these studies the socioeconomic status was in fact measured by income as it does here. There are some famous more recent studies, perhaps the best known as Michael Marmot and these colleagues work on point hall and the gradient in point hall where higher status civil servants and higher status in that case is actually measured by income, live longer than lower status civil servants and there's a bunch of comparable European studies. So these studies have been around for a really, really long time. I think they're important for all medical scientists or economists and for physicians and there's always been this tension which actually goes back into the 19th century between those who think of health as primarily a social phenomenon or a political phenomenon and those who think of health as a medical problem with a focus on disease, the focus on finding germs and you know it goes right back to the old debates between the miasmusists who thought the environment was what did it and the germ theory people who of course thought it was germs. In some ways I've been a very productive economy even though it's occasionally gotten quite bitter. I think what Rosh and his co-authors have done is really quite extraordinary. I mean it's sort of bringing this into the modern age of big data and basically taking You know, all of the numbers. we have the United States, the complete income tax records, the complete death files and putting them together. And that enables them with this enormous number of observations, about 1.3 billion person years to really nail these numbers in a very precise way that's never really been done before. And as Raj explained, you can also then look at the gradient, the gradient is the term that refers to those difference across income groups by mortality rates. And you can do it by particular areas too, as Raj said. But before we're getting into sort of more details, let me just say one, I think important qualification to what Raj said. He talked about life expectancy. When we talk about life expectancy, we usually refer to life expectancy of birth. And that's not what's being calculated here. Life expectancy being calculated is the expected rate age of death for someone aged 40. So it's more like life expectancy than 40. And that's important because a lot of the variations across space and across countries are to do with income mortality and child mortality, which has a huge effect on life expectancy, much more so than mortality rates after 40. And so that's really off the table here. And of course what happens to children is an important part of the gradient more generally. So I think that's an important thing to keep in mind when thinking about this study. This question really is for both of you. So most of our listeners will be physicians and most of our readers, people who read both the editorial as well as the paper will be physicians. And Raj, you mentioned it, the association between access to healthcare and some other factors in life expectancy at age 40 was not nearly as strong as certain health behaviors, smoking, obesity, and exercise. What do you think that means for a clinician and what do you think it means for the way in which we think about health and healthcare in the United States? So I think that fact needs to be interpreted carefully. What it's telling us is that area-level differences in life expectancy don't seem to be readily explained by measures of access to healthcare. So it's not literally that you aren't able to access a doctor or access quality medical care in these places where you're seeing lower life expectancy among the poor. Now that, of course, does not imply that healthcare doesn't have an important effect on life expectancy and health more generally because in the US as a whole, it's possible that it's just this area-level variation that's not that well explained by differences in healthcare access. Whereas for any given individual, if you are able to see a doctor when you have a life-threatening illness, of course, we think that that might have a causal effect on mortality rates. So I think what I take from it is that whatever channels we seek to explore and explaining these differences across areas and trying to improve life expectancy in some of these places, like some of the places with very low life expectancy, like Gary and Diana or Las Vegas and Nevada, is that the channel runs through health behavior most likely that whatever we do is ultimately going to have to influence the choices that people are making in terms of smoking and diet and exercise. And that could, in an important way, involve preventive healthcare and the healthcare system more generally. But I think it suggests that it's not literally about treatment of acute illnesses that's driving low life expectancy in some places. And guess what's your interpretation of these things? I agree with much of that. One has to be careful as far as it is by concluding that healthcare really doesn't matter. I mean, lots of people would like to think that. I certainly don't think that. And I think one of the criticisms of a lot of the social determinants of health literature does tend to minimize the importance of healthcare. I do, of course, agree that behaviors are terrifically important, especially the behaviors we talked about. One of the great unsolved riddles, and it's a point of contention, I think, between the more social and the more medical approaches, is just trying to understand why it is to put it in a perhaps-infolicitous phrase, the poor tend to behave poorly. You know, it's not like we haven't known smoking is bad for you for a very, very long time. And poor people know that as well as anything else, but they continue to smoke. And it may actually, in some people of our youth, it may actually be in their best interest to smoke, and given the alternatives and their lives, health is not everything. And people may be choosing to trade in some health from some other benefits they get from tobacco. Of course, they may not be, and there may be also sort of malign social forces, which are causing that to happen, but much the same is true of obesity and so on. So those things really are important, trying to understand why they happen, is something I don't think we really fully understand. Yeah. I think one of the big questions for me, though, if I can push on, and you can certainly stop me, is, you know, whether we really think income is a policy variable here, or whether we think income is really standing proxy for something else. And you know, that's not a question the paper takes up, nor should it. It's got plenty to say without dealing with these questions. But I think for a lot of people, and a lot of ordinary physicians is reading this, they'll say, well, you know, there's not a lot I can do. What I should do is tell these people, go out and get a good job, or tell them to get rich, and that'll do a lot more for their health than I can do or a physician can do. And I'm not at all sure that's true. I mean, I'm one of the people who's perhaps a little bit on the extreme of believing that income by itself is not that important, and that income is actually masking or confining a lot of other things like education. There's a huge amount of reverse causality, for instance, I mean many people who are very low income have been long-term disabled. The paper tries to deal with that a little bit, but you can't really deal with long-term sickness and it's a fact on income. There's a lot of stuff that happens in childhood that carries forward in health and childhood and education and childhood, both produce income in adulthood. And you know, so it doesn't necessarily tell us where it is. Education is terrifically important. I mean, I think one way of looking at these would be to disaggregate by cause of disease by sort of cause of death, which is something potentially it can be done with these data. And also, as I say in my editorial, I mean the education levels are on the death certificates over this period, so I hate to ask Raj to go and do more merging, given the most amazing merge she's already done, but it would be terrifically valuable to merge in both education and the cause of death into these reference. Raj, how do you respond to Angus's comments? So I completely agree with Angus's point that we shouldn't interpret the results in this paper as telling us anything about the causal effect of income on life expectancy. It's about the association between income and life expectancy. And as Angus notes, that association could well be, and I also agree likely to be driven by other factors like differences in education or just correlated differences in behavior that drive this association between income and life expectancy. There are two things in particular that lead me to believe that simply giving people more money is unlikely to lead to a commensurate gain in life expectancy, comparable to what we see in the associations we've documented. The first is we find an association between income and life expectancy throughout the income distribution, even at very high levels of income. You really thought it was about being able to afford health care or being able to afford the things that lead to a longer life. You would expect that relationship to diminish quite significantly at higher income levels. Yet we continue to find at income levels above $100,000 or $200,000 or even $500,000 that higher incomes are associated with higher life expectancy. So my view is it's quite unlikely that in that income range, it's literally the causal effect of having more money that's leading people to live longer. The second piece of evidence that leads me to think that these should really be interpreted as associations rather than causal effects is from another recent study that's just been published by Cessarini et ol, which looks at lottery winners and looks at how life expectancy varies when people randomly get more money. They find essentially no impact on mortality rates going forward consistent with the idea that the associations we're documenting are driven by some third factor, not literally the effect of having more money itself. So what I take from that is, again, we need to understand what it is that's making the poor behave more helpfully in certain circumstances. And I think the spatial variation that we're seeing here provides useful clues. So it's not that it's just immutable that low socioeconomic status individuals are more likely to smoke or more likely to be obese. In fact, we find that in certain places in America, like San Francisco, in New York, The poor actually seem to behave much more helpily in these affluent educated cities and correspondingly they live quite a bit longer. So I think understanding what's going on there can give us a more concrete angle to figuring out these issues. And I absolutely agree going forward if we can look specifically at causes of death and look at the mediating role of education we'll have a further handle on what's going on. And just to add a word on there. I actually think one of the most fascinating things through this study is that something that no one's done at all before is to show that this grade is really there at the very, very top. I mean, I'm not sure if most of us have been asked whether we would have known the answer to that because most nearly all the studies before you're not really looking at the super-rich here. And it's a really curious question. I mean, what is it about the super-rich? You know, the cost system to live longer if it's not income. And there's a wonderful bug in which people looked at billionaires who'd lost their billions and then got it all back again and asked them what about them being a billionaire was the best thing. And they all said the same thing. It was the private jet. So, you know, maybe it's the private jet that's keeping people healthy. But who knows? I'm always skeptical of lottery because lottery is not like getting more money in your job. And, you know, there's a long literature in psychology and elsewhere that lottery winners often have a pretty miserable time of it. And, you know, winning a lottery is not just like getting more money. But let me come back to something that I think is really, really important and which I think it's really vital to look at what's happening over time as well as to look at what's happening over people. And that's where, I think, healthcare really kicks in and is really important. In a paper that I wrote with one of Roger Kool-Offers, David Cutler, we have this sort of split which says the major declines in mortality of the last 50 years come from, you know, three sources. I mean, one is reduction in smoking. One is better treatment for cardiovascular disease, particularly anti-hypertensives and that sort of thing. And the third one is all these other things think together that we don't really know. You don't get anti-hypertensives in the pharmacy. You've got to go check with your doctor and that's something incredibly important in positions that are done to help extend our life expectancy over the last 50 years. And, you know, you're not really going to see that by looking across space, for example, but it's nevertheless incredibly important. I guess you had commented on some of the data that you found fascinating. I was going to ask you as you read through the manuscript, what else struck you as a surprise? This has been your area of investigation for many years. What else struck you as being surprising or unique? Well, I think the difference in the spatial gradient are surprising and unique. And I'm not quite sure what to make of them. And one thing that I would be concerned about there is that the ethnic and racial mix is different in different cities. And it's not easy to know exactly how to adjust for that. And it's possible that on further investigation, some of that will be attributable to differences in the way that race and ethnicity work in different cities. And I think that is potentially important. There's some work showing that the racial makeup of the city, for instance, is important and the health outcomes not just for those races, but for other groups. And that means the patterns across race and ethnicity are likely to be different in different places. And you can't do everything in one study. And so these are certainly intriguing. And there are things that would be really nice to look at. I mean, the other thing going forward that would be really very interesting is, when you calculate life expectancy, you hold mortality rates constant. And that's just the way life expectancy is calculated. I mean, what you're really looking at here is variation in mortality rates by age, by income group or income per centile. And what you'd really like to do is take a 40-year-old and say what are the various mortality rates that are going to face at various ages going forward. And also if that 40-year-old is in the 100th percentile now or the 20th percentile now, which percentiles are they going to be as they go forward? They're not going to stay in the same percentile, nor are the mortality rates going to stay the same as we go forward. And that, again, is something that can't be tackled here, but is going to have to be tackled as we move forward. So I completely agree with Angus that some of these differences we're seeing across areas could play out in the way that mortality differs across races and ethnic groups. I do want to just clarify for listeners that the estimates we're reporting do attempt to adjust for differences in racial and ethnic composition across areas. So the estimates reflect the level of life expectancy we would see if every area had the same fraction of whites, blacks, Hispanics, and Asians as the US as a whole. However, our methodology assumes that there are constant differences across those groups in rates of mortality regardless of space and what Angus notes, which I agree with, is that it's possible that upon further examination, we'd find that in certain places there are smaller differences in mortality between, say, blacks and whites or Hispanics and whites, and that could explain some of the local variation that we're seeing. But at a big picture level, it's important to note that we're not just seeing lower levels of life expectancy in certain places because there's a larger African-American population. We're making an effort to adjust for that at a first-order level. The other point I wanted to make is about changing mortality rates across ages. So as Angus notes, we're looking at how your income percentile age 40 predicts mortality rates at later ages, which we then translate into estimates of life expectancy. It would be very interesting to look at how income changes over time and see how that's associated with changes in mortality rates. One result that's somewhat interesting is if we instead look at mortality rates as a function of income at age 35 or as a function of income 10 years ago instead of income two years ago, which is what we do in our baseline analysis, we find pretty similar patterns. So the reason for that is people's incomes while they fluctuate over their lifetimes. They actually are overall quite stable once they reach age 35 or 40, which is part of the reason we start our analysis at age 40 rather than looking at mortality rates at earlier ages. And so we've done some investigation of varying the ages at which we're measuring income, and we tend to find pretty similar mortality profiles. So I think, again, it would be interesting to investigate this further, but there's something about income at any given age that seems to be associated with people's behaviors that then predicts the full sequence of later mortality rates. Raj, just one more question for you and then Angus, a question for you. Raj, you've lived with these data for a year, and it's quite a few data points. We certainly ask for a few revisions. There's an incredible amount of rich supplemental data. When you think back and you think about where you started, what's been the biggest surprise in the data? I think at the end of the day, I was most surprised by the differences and the trends in life expectancy across areas, and what is associated with these differences in life expectancy across areas. So going into this project, I had some sense from earlier work that we would find big differences in life expectancy in the nation as a whole, that we would find divergent trends, although the exact shape of the gradient is Angus noted, especially at the top of the income distribution was unclear. But the spatial variation and how it would turn out, I think, was not something that any of us really had a strong prior view about. And so to give you one example in other work that I've done on differences in intergenerational mobility across space, and in a number of other studies, you often find that the south looks the worst in terms of economic, social, and health outcomes. And interestingly here, once you focus on the low income population, that actually doesn't turn out to be the case. The south does not have unusually low rates of life expectancy for the poor. It's roughly close to the average, and instead, it's places in the rust belt that have very low levels of life expectancy and often declining life expectancy over the past 10 or 15 years. So I think the spatial patterns surprised us and the magnitude of the spatial variation surprised us. And then seeing that this really seems to be strongly associated with health behaviors and there's some positive effect associated with being in these affluent educated cities for the poor, those were not results we had anticipated going in. We don't have clear policy conclusions yet of what to make of that. But what we can say, I think, is that we should really be thinking not just at a national level, but also at a local level and thinking about how to address issues of life expectancy and health because there really are very big differences across places. And we should be thinking about how to change health behaviors in those places like Gary and Diana, for example, for example, focusing on the course. I'm hopeful. that some of those insights will be useful for policy and practice going forward. Thanks, Raj. And Angus, the last question to you, this past fall, you were the recipient of the 2015 Nobel Prize in Economics. What was it like? It was wonderful. I mean, that's, I suppose, the simplest thing. It is a little bit like I've been comparing it to, you know, this is the old story, but the dog that spends his life chasing buses, and then, you know, one day he has the turbo in his portion to actually catch a bus. It has no idea what to do with it. And of course, for me, this was not a turbo in his portion. This was a wonderful fortune. But there is a sense of being run over by a bus, and it takes away all your time. I mean, it's like sucking all the air out of the room, it sucks all the time out of your life. On the other hand, there were episodes of just surpassing wonderfulness. The week in Stockholm is something that is quite unparalleled, and it is a truly extraordinary glittering, wonderful time. And I wouldn't have missed it for that, I think, it's truly wonderful. Well, Angus, I can't thank you enough. I'm sure my email to you in January about the paper was buried amongst many thousands of others' emails. But you managed to see John in the subject line, and you were kind enough to answer it. So I do want to thank you. And Raj, my career, before I came to John, was based at the Old Boston City Hospital, Boston Medical Center. And my clinical life was really amongst marginalized and poor populations. And so, this paper was very important to me personally, but I really want to thank you one for doing the research, the extraordinary research. And then secondly, of course, I want to thank you for sending it to JAMA. It's a real privilege to be able to publish the paper today. This is Howard Bachner. I want to thank both Angus Deaton from Princeton and Raj Chetty from Stanford for joining us. I urge you to read the editorial, read the paper, and in addition, there's two other editorials that will be accompanying the paper both in print and online today. Thanks again for listening. Links to the article and editorial are in the show notes in iTunes. If you enjoyed this podcast and would like to hear more, go to iTunes.com/jama-network and subscribe. While you're there, check out our entire roster of podcasts. All of our podcasts are also available in Stitcher.