Measuring the Long-Term Health Effects of Air Pollution, with Tatyana Deryugina

Notable Quotes

• Air pollution can accelerate aging: “Air pollution damages your body, damages your cells. Some interesting correlational research has found that people who have more lifetime exposure to air pollution have shorter telomeres on their DNA … So, we even see this correlation at the cellular level, where it looks like air pollution is causing damage that basically speeds up the process of aging.” (6:59)
• Reduced exposure to pollution offers long-term individual benefits: “Early in life, the mortality impacts are trivial … because most young individuals have very high levels of health, and so changing their exposure by a little bit is not going to kill them when they’re in their 20s or 30s. But, as those individuals age, that little extra health capital from having less pollution [exposure] starts making a big difference. In fact, we estimate that most of the survival benefits accrue to those over the age of 65 … and about 90 percent for those over the age of 50.” (27:47)
• Continued large-scale benefits from reductions in air pollution: “People talk about what the mortality impacts are when pollution goes up by one unit, but in fact, we’ve had huge reductions in air pollution in the United States: [sulfur dioxide] is the largest in terms of the shares, but other pollutants, as well. I think that is really good news—that the air is much cleaner now than it was in the 1970s, and our estimates do suggest that the benefits are large and that we haven’t even seen them all yet.” (30:17)

Top of the Stack

• “The Mortality and Medical Costs of Air Pollution: Evidence from Changes in Wind Direction” by Tatyana Deryugina, Garth Heutel, Nolan H. Miller, David Molitor, and Julian Reif
• Lessons from the Edge: A Memoir by Marie Yovanovitch

The Full Transcript

Kristin Hayes: Hello, and welcome to Resources Radio, a weekly podcast from Resources for the Future. I’m your host, Kristin Hayes. My guest today is Tatyana Deryugina, associate professor of finance and Shebik Faculty Fellow at the Gies School of Business at the University of Illinois Urbana-Champaign (the main University of Illinois campus). Today, we’re going to be talking about new research that Tatyana and her coauthor, Julian Reif, have been undertaking on the impacts of chronic air-pollution exposure on life expectancy.

They’ve been pioneering some new techniques that allow for better estimations of the long-run impacts compared with the short-run impacts that are often reported by economists. The pair’s research offers what I would argue are some sobering findings, as well as one key uplifting one—and, for those of you who love the “research weeds,” we’re also going to talk about some innovations in methodology today. Stay with us.

Hi, Tatyana. It’s really nice to be here with you in person in the Resources for the Future (RFF) office today. Thanks for coming on Resources Radio while you’re visiting RFF.

Tatyana Deryugina: Thank you very much for having me.

Kristin Hayes: Well, you are an economist by training, but you’re now at a business school, and I noted that you’re an associate professor of finance. Can you say a little bit more about your research interests and how you ended up where you are today?

Tatyana Deryugina: Sure. My research interests broadly fall into the field of environmental economics. Much of my work has been on the economic impacts of natural disasters and the health effects of air pollution. Our finance department actually has a group of applied economists called the Center for Business and Public Policy, and a lot of the research that we’ve done there does focus on, I would say, non-finance topics, and that is by design. I joined explicitly as a non-finance person recruited by a finance department.

Kristin Hayes: Interesting. Does that mean there’s a cohort of economists there with you? Is this something that the university has really carved out as a specialty within its business school?

Tatyana Deryugina: Yes. It wasn’t so much university-driven but originated from the finance department, because some of the members were doing less traditional finance topics and the finance department wanted to expand on that. My coauthor, Julian Reif, is also there, and Nolan Miller and David Molitor have been there for a long time. Don Fullerton just retired—he’s a well-known environmental economist—and we had a new person, Mackenzie Alston, join us last year. We’re a small but mighty group.

Kristin Hayes: Very cool. Well, the piece of research that we’re going to be talking about today focuses on, again, the long-run impacts of chronic air pollution on mortality, and in particular on comparing estimations of those long-run impacts with what I believe are more typical analyses of shorter-term impacts.

There’s going to be a lot to unpack here today—but, again, this difference between long-run impacts, short-run impacts, different types of analyses. Let me start by asking you—why were you and Julian keen to explore this research question overall?

Tatyana Deryugina: Over the course of my research, I encountered a lot of reports from the World Health Organization and similar organizations talking about how many people are killed by pollution each year—prematurely killed. When you look at the sources for those numbers, they’re usually from epidemiological studies that correlate life expectancy with ambient pollution levels.

Now, these studies are aware of the potential biases in interpreting simple correlations as causal, so they try to control for a large number of potential confounders. But economists have long realized that there’s only so much you can do in accounting for variables. At some point, you run into the problem of things that you cannot observe, and that’s when you need to bring in a different methodology, which economists have done, going back over 20 years now, with the first what we call “quasi-experimental” studies.

But these studies have a major limitation, in that they typically focus on health outcomes of one year or less—sometimes a month, sometimes a week, sometimes even the same day. It made us wonder if this is one reason why we still see the epidemiological estimates being used, because, while they might have a lot of flaws methodologically, the quasi-experimental research done by economists has another major flaw, and that is not being able to consider things like life expectancy, but just looking at mortality over one year.

Kristin Hayes: The ranges that you talked about there are considered short term—that one month, even one day, and all the way up to a year is considered a short-term impact or a short-run analysis for this purpose? Is that right?

Tatyana Deryugina: Yeah. I would consider them short-term outcomes in the sense that they cannot be used to say by how much someone’s life was cut short. I mean, I don’t think there’s a magic interval where something goes from short run to long run. It’s a continuum. But, in general, just having one year’s worth of outcomes is probably not going to be good enough for the World Health Organization.

Kristin Hayes: Got it. Yeah, that makes a lot of sense.

I want to talk a little bit about another important contextual piece of this, which is about the different mortality pathways that air pollution can lead to. Some of the numbers that you cite in the paper are pretty stark—pretty sobering, to use that word again—about the scale of the problem and show that there are really significant impacts from air pollution on morbidity and mortality, but what are the actual pathways that we’re talking about there?

You mentioned a few of those in the paper. You mentioned that air pollution can, for example, lead to hardened arteries in otherwise healthy individuals, and that can, in turn, lead to heart disease, so that’s one pathway. I will note that that type of impact—I believe the term that’s in the paper is “accelerated aging”—falls into the category of things that happen to people, but air pollution makes them happen faster. Is that right?

Tatyana Deryugina: Right. That’s one way to think about it. Air pollution damages your body, damages your cells. Some interesting correlational research has found that people who have more lifetime exposure to air pollution have shorter telomeres on their DNA. Every time our cells replicate, the telomeres shorten. They’re at the end of chromosomes. The shorter the telomeres are, that’s considered a marker of aging. So, we see this correlation at the cellular level, even, where it looks like air pollution is causing damage that, yes, basically speeds up the process of aging.

Kristin Hayes: Okay. Very, again, interesting and sobering. Then, there’s another type of impact—I think it’s fair to categorize this as a different type—where air pollution can affect, let’s say, the respiratory and cardiovascular systems of people who might already be very frail, and that represents a different kind of impact. Can you say a little bit more about the distinctions between those types and why that distinction matters? To put that all in context.

Tatyana Deryugina: Yeah, and I should say that the second type of impact—where it’s really the already-frail individuals who are being affected through their respiratory system, through their cardiovascular system—is what gives rise to some of the skepticism surrounding these studies that focus on short-term outcomes. A year or less is, well—how long would these individuals have lived?

This is not to say that prolonging someone’s life by a year is not valuable, but it really matters if the person would’ve lived for five more years, had air pollution not killed them, or if they would’ve died a few days later. This kind of phenomenon is called “mortality displacement,” where something kills people who would’ve died very soon otherwise.

It happens not just with pollution—or it’s thought to happen not just with pollution—but with heat waves, for example. The individuals that die as a result of heat waves tend to be pretty frail. The obvious air-pollution deaths also tend to happen to pretty frail people, if we’re talking about acute, very short-run exposure to extra air pollution.

Kristin Hayes: For the purposes of this study, you were particularly looking at SO2, is that right—sulfur dioxide? That was the pollutant that you were looking at most closely?

Tatyana Deryugina: Yes. Sulfur dioxide is a pollutant that’s emitted by coal-fired power plants, mostly. During our sample period, which is 1972 to 1988, it was the most widely monitored pollutant in the United States. It was also much more prevalent than it is today. Since the 1970s, SO2 concentrations have dropped by over 90 percent in the United States, so we’ve really cleaned up coal-fired power plants, partly by getting rid of them, and partly by just having them work less, have pollution control equipment, and so on.

But yes, that’s what we focus on, partly because fine particulate matter—which is another important pollutant that is of big concern today—was not monitored in the 1970s, and also because SO2 converts to a subset of fine particulate matter called sulfate. Our estimates are relevant for both thinking about SO2 and for thinking about fine particulate matter, because it’s a natural conversion process that will just happen as the SO2 travels through the air.

Kristin Hayes: Okay. I want to go back to this short-run/longer-run question for just a second, and you’ve mentioned a couple of the flaws that are inherent in the way that these short-run analyses are carried out. Why has it typically been more challenging—knowing these flaws and knowing the benefits of estimating longer-run impacts—to do that historically?

Tatyana Deryugina: Quasi-experimental methods require there to be some variation that’s as good as random. In this case, it has to be variation in pollution that’s as good as random, so economists have identified a variety of such settings. For example, there’s a paper that uses the introduction of E‑ZPass, which allowed cars to pass through toll roads more quickly, reducing emissions and lowering air pollution for people who lived around the road. There are papers that use the NO2 (nitrogen dioxide) trading program because there were some changes in that they’re able to exploit, and so on.

In our paper, we use changes in wind direction—day-to-day changes. It’s relatively straightforward to find these as-good-as-random short-run changes in air pollution, a lot harder to find longer-run differences in air pollution that are as good as random. And, even if you find them, if people respond to them, then it’s going to get harder and harder to interpret your estimates.

Suppose pollution goes up and all the sick people move away. Well, now you have to first follow those people over time to make sure you’re still tracking the right cohort of people. Then, you also have to try to account for the fact that they moved, since now their pollution exposure has changed, and maybe other things in their lives have changed. And whatever mortality number you get out of that is going to be potentially really, really complicated to interpret.

Kristin Hayes: Okay. Well, this is great. I promised research weeds, and we’re already wading into them just a little bit, but I want to continue our talk about methodology. You and Julian note—I’m going to quote from the paper here—that “the main contribution of our study is the development and application of a new framework for estimating the long-run survival effects of chronic exposure to environmental hazards.”

Again, let’s talk methodology for a bit, and given that seminal contribution here, I wondered if you could talk me through the stages of your research. How do these short-run analyses looking at short-run mortality impacts move ahead to where you’re able to look at more chronic exposure and these longer-run mortality impacts?

Tatyana Deryugina: Yeah, I’m happy to. The first, roughly, half of our paper is very similar to what a typical quasi-experimental study would be: we use short-run variation in air pollution with wind direction as a source of variation, which is very nice and intuitive, and then we estimate the short-run mortality consequences of that. We’re able to look at mortality up to a month following exposure, and then it just gets statistically more difficult to tease out the signal from everything else that’s going on. So, that is pretty standard.

What we do that’s really different is we take a model of survival that was recently developed by Lleras-Muney and Moreau that’s basically a model of human survival that can incorporate a variety of dynamics into it. If you’ve ever seen a survival curve, initially there’s high mortality at the very beginning of life. Infant mortality is relatively high, then mortality is very low for the next 20 years of life, at least in the United States. Then, it picks up a little bit over the next 20 years. Then, eventually, it picks up a lot more, so you get this curve.

This model can be used to represent typical human mortality curves with several parameters that are calibrated, and our innovation is that we basically take our short-run estimates and use them to inform the parameters of the model. It has several parameters, like alpha, delta, epsilon, sigma. There’re a couple of sigmas in there.

We calibrated to the 1972 mortality profile to imagine a cohort that’s born in 1972 to start with. Then, we look at how we see mortality change at different ages as a result of one-day SO2 exposure, and we use those changes to inform how the parameters of the model must be changing, and that allows us to have a structural model of how SO2 exposure affects your health.

Ultimately, the model is one where there’s latent health that changes every period. When your health gets low enough, you die, so we infer what the underlying parameter dynamics are to create the mortality effects that we actually see in the data. Does that make sense?

Kristin Hayes: Yeah. This makes me want to ask you a bit more about the data sources that you brought together, as well, because it does sound like there were some very specific choices made around the use of wind-direction data, for example. Also, I want to ask a little bit more about this 1972-to-1988 time period. You mentioned that that’s a time when SO2 was quite high, but are there other reasons that you focused in on that time period related to data availability? Maybe this is a moment where I can just ask you to speak a little bit more to the data that you brought to bear as you were actually calibrating this model.

Tatyana Deryugina: Sure. Using wind direction as a source of experimental variation in air pollution is actually something we’ve done in a previously published paper in a different time period with a different pollutant, and we showed that it worked quite well there. The idea is really simple—you just look at what direction the wind is blowing from and you relate that to the pollution levels, controlling for some basic things like the average pollution levels in the county and some things called fixed effects, if our listeners are familiar with those.

The nice thing about using this kind of variation is it’s plausibly random and it’s not data heavy, because you can imagine doing an atmospheric transport simulation where you know where all the coal-fired power plants are and you know how much they emit, and you have all the weather patterns, including temperature, and you model the atmospheric chemistry, and then you predict where pollution gets blown around. That can also be a good source of as-good-as-random variation, but it’s very computationally intensive and has fairly high data requirements, because you need to know emissions, and you need to know atmospheric conditions in detail.

In our case, going back to the 1970s, there are no detailed atmospheric conditions—at least not on a granular enough spatial level, but we do have so-called “re-analysis data,” where several data sets are combined together to infer something about atmospheric conditions that none of the data sets alone can do as good of a job reproducing. So, we take these wind-direction re-analysis data, and they give us broad regional wind patterns. Then we do need pollution data, so luckily, the US Environmental Protection Agency was monitoring SO2 back then, so we bring that in, as well. Then, we also have county-level daily mortality data by age, race, and sex, although we only break it down by age; we don’t break it down further.

That is really the limiting factor in terms of the years spanned by our dataset, because, for some reason, the CDC (the US Centers for Disease Control and Prevention) decided that, after 1989, you can no longer have daily data at the county level publicly available. You would have to apply for restricted access through a research data center of the US Census Bureau and work with the data there, which we decided not to pursue at this point.

Kristin Hayes: Interesting. But that data from 1972 to 1988 is, in fact, still publicly available?

Tatyana Deryugina: Yes.

Kristin Hayes: Oh, that’s interesting.

Tatyana Deryugina: It is publicly available. I guess, maybe because it’s a long time ago—they decided a sufficiently long time ago—that it’s okay to make it publicly available. Maybe there are fewer other public records that you could use to identify people. I’m not really sure, but yes, it is out there.

Kristin Hayes: Okay. Well, I have asked you a lot about methodology and data and innovations. I feel like I always do this to our poor listeners—I leave the juicy stuff to the very end—but, of course, I want to talk about your findings, as well.

Just to reiterate, you started with this acute exposure analysis, so let’s begin there. Let’s begin with, as you mentioned, that first part of your paper, which looks at that short-run analysis. What can you say about what you were finding as you were combining these data sources together?

Tatyana Deryugina: I can give you the numbers, which is that, when SO2 goes up by one part per billion—which is about 10 percent of the mean in our sample—mortality goes up by about 0.08 deaths per million on that same day.

Now, that shouldn’t mean anything to you, because nobody thinks about daily mortality per million people, but I think a key interesting fact is that we find this one effect for one-day mortality, and then we look at longer-run mortality, still holding the change in SO2 constant. So, the change in SO2 is always one-day change by one part per billion.

A couple of things can happen, theoretically. First, as you extend the time window, if the individuals who are dying are those that are very frail and would’ve died soon anyway, you will actually see a declining pattern of your estimated effect. So, if all those individuals would’ve died within a week, then, when you extend the follow-up window to two weeks, you will find a zero, because over that period of time there were no extra deaths caused by pollution. All those people would’ve died anyway. This is where I like to make a joke that the 200-year effect of pollution on mortality is zero.

Kristin Hayes: Is zero for everyone.

Tatyana Deryugina: Yes, it’s zero. You look out 200 years: zero impact of pollution on mortality. That’s why it’s really important to get the right time frame.

The other thing you could find is that the effect grows, and that would suggest that there is this accelerated aging happening, that there’s this lagged effect of past exposure.

Remarkably, if you think about our estimates being at the daily level, we do see the effect growing with time, so that the monthly impact of this one-day shock is about twice as large as the one-day impact, which suggests that some people are not being killed immediately but are dying as a result of this exposure a week, two weeks, three weeks later.

When we break it down by cause, we also see something interesting. When we look at cardiovascular causes, they behave in the same way as our overall estimate—they grow with time. When we look at cancer mortality in the short run, very short run, on the same day, it’s responsible for about a third of the extra deaths.

But then, when you go out about two weeks, the coefficient drops to zero and becomes statistically insignificant and stays insignificant, which is basically saying that SO2 is killing people who are already very sick, because you obviously can’t develop cancer and die from it in one day as a result of pollution exposure. So, these individuals already had cancer, and their deaths were not recorded as being due to pollution, because that’s obviously very hard to trace, but they were very frail. We find both of those patterns.

We also find similar distinct patterns by age. When we look at individuals who are 60 and older, we see an increasing pattern over time, suggesting that lagged effects are prevalent for this population. When we look at younger individuals, we don’t find any effects for anybody 20 and under, but for people between 20 and 59, we see one-day impacts, but then they go to zero over a month, suggesting that there are some younger people that are killed by air pollution, but they’re very sick young people; they have very little life expectancy left, in terms of the acute effect, of course.

Kristin Hayes: Okay. Let’s turn to the longer-run innovations and the new approach that you took. We talked a little bit about the model—what does this model that you’ve calibrated show about the effect of chronic exposure on life expectancy compared to what you would’ve guessed if you just took those short-run estimates that you were talking about and extrapolated them? You’ve got this limited viewpoint that you can extrapolate for the long term versus the more thorough long-term estimates that you guys are doing with the model. How do those compare?

Tatyana Deryugina: What we call the “naïve” estimates are just a linear scaling of our monthly impacts. You calculate what is the effect of each age, you scale it up by 365 days if you want a year exposure, more if you want lifetime exposure, and then you integrate over the age distribution to figure out the life-expectancy estimates. You get one number that way.

When we use our model, which basically takes into account the fact that some individuals are not immediately killed by pollution but become sicker, lose some of their health as a result, and are then more likely to die later, we get much bigger estimates for the older ages. The naïve model actually overshoots at younger ages—those below 55 or so—and then predicts much lower life-expectancy impacts at older ages compared to our model of chronic exposure.

It’s because when we use our model for chronic exposure, initially, in terms of early in life, the mortality impacts are trivial. They’re not really there, because most young individuals have very high levels of health, and so changing their exposure by a little bit is not going to kill them when they’re in their 20s or 30s. But, as those individuals age, that little extra health capital from having less pollution starts making a big difference. In fact, we estimate that most of the survival benefits accrue to those over the age of 65—it’s about three-quarters—and about 90 percent for those over the age of 50.

Kristin Hayes: Very interesting. So, everyone over age 50 really needs to move to someplace with very good air quality, right, because the benefits accrue to them at a disproportionate amount compared to the population overall?

Tatyana Deryugina: Well, but our model actually has chronic exposure in it, so the counterfactual that we have is everybody is exposed to one unit less of SO2, but those benefits don’t materialize until age 50. It’s kind of like, if you’re a smoker, you’re not going to get lung cancer in your 30s, but the fact that you smoked in your 30s might lead to lung cancer when you’re in your late 50s and 60s. That’s kind of what’s happening here, too. It’s not that people should only care about their exposure when they’re old, it’s that the consequences of early-life exposure happen at much older ages.

Kristin Hayes: Interesting. Well, I did promise our listeners one uplifting finding, and this is how I interpreted it, and it’s very much related to what we’re talking about right now in terms of the long-term benefits. I’ll just quote, again, from the paper, where you noted that “90 percent of the survival benefits accrue after the first 50 years of life,” as you just mentioned, “implying that most of the 1970 Clean Air Act’s health benefits have yet to emerge for cohorts born after its passage.”

I did take that as somewhat of an optimistic note, which is that the flip side of those negative effects showing up much later is that the benefits also show up much later. Can you help us interpret that comment just a little bit? Am I fair to think that that is kind of a positive story coming out of this?

Tatyana Deryugina: Typically, people talk about what the mortality impacts are when pollution goes up by one unit, but in fact, we’ve had huge reductions in air pollution in the United States: SO2 is the largest in terms of the shares, but other pollutants, as well. I think that is really good news—that the air is much cleaner now than it was in the 1970s, and our estimates do suggest that the benefits are large and that we haven’t even seen them all yet.

Now, here I will make an economist’s caveat, which is that our estimates should be interpreted as the benefits that would materialize with no behavioral change, because we’re using these short-run shocks in air pollution that people are unlikely to respond to by moving, and so on. We think this is very policy relevant, because it tells you what the effects would be if people took no action.

But it’s different from predicting—we’re not predicting what US life expectancy will go up by, but what would it go up by had people not done anything in response, because, obviously, we’ve got a lot of other things going on right now. I don’t want to say that we’re making very specific predictions about where life expectancy’s going, but we are estimating very substantial benefits of the reduction in air pollution over the past 50 years.

Kristin Hayes: Fantastic. Well, this has been great. I really appreciate your taking the time. You’ve done a fantastic job talking us through the methodology. I know there is a significant subset of our audience that really loves this stuff, too—fixed effects will be music to ears for some of our listeners out there.

Tatyana Deryugina: Great.

Kristin Hayes: Let me close with our regular feature, Top of the Stack. We’re looking for recommendations of other good content that you might want to suggest to our listeners. Feel free to let your imagination run wild. What’s on the top of your stack?

Tatyana Deryugina: As you know, I’m from Ukraine originally, so that’s been taking up a lot of my head space and time over the past year and a half. One book that I’m reading right now is Marie Yovanovitch’s autobiography, Lessons from the Edge. She was an ambassador to Ukraine during the Trump administration, and she was involved in some of the scandals surrounding Ukraine—in a positive way, I should say. She was, I would say, the hero in that situation. I had the privilege of meeting her in the past year, and her autobiography is just fantastic, and that’s what I would recommend to the listeners.

Kristin Hayes: That’s a great recommendation. We’ll post a link to it. Thank you again. I know this is the very beginning of your visit to RFF, so I personally will look forward to seeing you over the next few days. Thank you for starting off here at Resources Radio.

Tatyana Deryugina: My pleasure. Happy to be here.

Kristin Hayes: You’ve been listening to Resources Radio, a podcast from Resources for the Future.

If you have a minute, we’d really appreciate you leaving us a rating or a comment on your podcast platform of choice. Also, feel free to send us your suggestions for future episodes. This podcast is made possible with the generous financial support of our listeners. You can help us continue producing these kinds of discussions on the topics that you care about by making a donation to Resources for the Future online at rff.org/donate.

RFF is an independent, nonprofit research institution in Washington, DC. Our mission is to improve environmental, energy, and natural resource decisions through impartial economic research and policy engagement. The views expressed on this podcast are solely those of the podcast guests and may differ from those of RFF experts, its officers, or its directors. RFF does not take positions on specific legislative proposals.

Resources Radio is produced by Elizabeth Wason, with music by Daniel Raimi. Join us next week for another episode.