Episode 004: Kevin Milligan

Economics

Oct 21

Prof. Kevin Milligan is an economist at the University of British Columbia. He works on economic topics that relate to everyday life, like universal childcare, income inequality, and tax fairness. You can follow him on Twitter @kevinmilligan.

Transcript

Cameron Graham: My guest today is Prof. Kevin Milligan of the University of British Columbia. His research covers a wide range of topics that relate to everyday life: universal childcare, parental leave, child tax credits, Social Security benefits, education, tax fairness, income inequality, retirement savings programs, and basic income. He is a fellow of the CD Howe Institute, a columnist for the Globe and Mail, a writer for Maclean's magazine, and the thumbs behind a prolific Twitter account. Prof. Milligan spoke with me from Vancouver in June. I hope you enjoy our conversation. Kevin, thank you for joining us on the podcast!

Kevin Milligan: Thanks for having me on. I’m happy to be here.

Cameron: UBC is renowned for its economics faculty. I think it's quite indicative of the increasingly dominant position economics has taken in the academic world that your economics faculty now has a very prestigious address, at 6000 Iona Dr. It’s a gorgeous old building, and I know because that's where I went to seminary way back in the 1980s. The Vancouver School of Theology has been displaced by the Vancouver School of Economics. I think there must be some kind of an object lesson in there somewhere.

Kevin: Yeah, well, I wouldn't say “displaced.” I mean, they just moved across the courtyard into a beautiful new building. They weren't making full use of this beautiful building and so they were able to put some money away and take account of their financial affairs, and ended up in a nice building, and we ended up with this beautiful pile. And we're very happy to be here.

Cameron: Yeah, it's a really interesting building. There's a couple of things I want to talk to you about. Your research is the main thing. I want to dig into one of your papers that you and I have had a chance to correspond about, and then I want to talk to about how you communicate about that research. Does that sound like a plan for you today?

Kevin: I'm good with it.

Cameron: Right, so let's start with your general choice of topic areas. You deal with a lot of things that touch people in their everyday lives, like healthcare and education, mat leave or parental leave, whatever you want to call it. What is it for you that draws you to this kind of topic?

Kevin: My fields in economics are what we call public finance and labour economics. And so I'm often drawn to the intersection of how different public policies and tax policies and spending policies affect people in the labour market, and also people in their family life as well. So that's a general area of interest. Two ways of thinking about what I'm trying to do: One is looking at data and looking at numbers allows us to bring evidence on different theories of how labour markets work, how families work, how they make decisions and what goes on. It allows us to build better theoretical models. And then the second arm of that is that by evaluating the impact of different public policies, you can have some influence on the conversation about what are effective public policies, and what they're trying to do, what they can do, and what they do do.

Cameron: Hmm. I'm used to, when I talk to economists, that there is this big distinction between macro and micro economics. Is that a useful distinction to you, or are you just wading in wherever you need to look?

Kevin: I am clearly a micro economist. We refer to what I do as I applied micro, these days, which generally means looking at individual date on people or firms or families, and trying to see how they respond to the incentives and the environment they have around them.

Cameron: So the paper you and I agreed to talk about is one on longevity. It's a topic I think that draws together a whole bunch of threads of a social policy, because every social policy is going to influence, ultimately, how long somebody lives.

Kevin: Yeah, the starting place for understanding what we're trying to do in this paper is that overall life expectancy has increased a lot. It really start taking off in the 1960s in Canada, and over that time period, from 1960s until, oh, the latest aggregate data we have is like 2014 or so. There's an expansion of about five years of extra life after age 65. It's a really large increase. That's for men, and it's around the same for women. Interestingly, two years of that have come just over the last 10. And so that's just a tremendous increase in life expectancy, on average. That was one piece of the starting point. The other piece is, in the US, researchers have found that that life expectancy change has been lopsided. While on average, it might be very large, turns out that people who were in the bottom half of the income distribution saw almost no change to the life expectancy, while people in the top saw big changes. And as you said in your introductory comments, you think about inequalities of various aspects of our lives, it's hard to think of a more fundamental one than the number of sunrises and sunsets you get to see. And if life expectancy is expanding at a very differential rate across the population, that's something we ought to know about. So that's why Tammy and I decided to embark on on this research project.

Cameron: Tammy is your co-author, Tammy Schirle?

Kevin: Yeah, Tammy Schirle is a professor of economics at Wilfrid Laurier University.

Cameron: When did you first become aware of this as it is a potential topic, this gap between rich and poor when it comes to life expectancy?

Kevin: It was somewhere in 2013 or '14, I started thinking about it. I was fairly freshly tenured at that point, I was just a few years into being tenured, and I decided I wanted to try to take some bigger risks, ask some bigger, more fundamental questions, and take on some longer-term research projects. And rather than looking at data this in front of me and try to figure out what to do with it, I decided to start with some questions I thought were interesting and seeing what data might allow me to answer those questions. I started to think about this question of mortality and and longevity, and started thinking about what would be the right data to allow me to actually look at that. I started thinking about whether it would be possible to get access, as a researcher, to the Canada Pension Plan records. Starting in 2013, I called up an old friend who was working at ESDC and got some tips on who I might talk to there. To make a long story short, in the summer of 2017, we had signed a contract to begin research on the data. It was a long and winding road to get there, but were able to get access to anonymized, de-identified data from the Canada Pension Plan administered records. That allowed us access both to people's entire earnings histories -- so all their earnings back to 1966 -- and also their longevity, because we could see when people died. And relating those two things together allowed us to study the impact of changes in longevity across different earnings groups.

Cameron: Okay, so just to clarify, then, what you got here, what you negotiated with the CPP, is that they prepare a database for you where they take all of their records and they strip out the personal information, but they leave some kind of a unique identifier so you can tell which records go with which individual?

Kevin: Exactly. So they make up a new pseudo-identifier, so we don't see anyone's social insurance number, not their name or anything like that that could identify you. And so there's a new pseudo-identifier attached to it. That then allows us to put together your earnings history with your benefit history, with your mortality record.

Cameron: Okay

Kevin: And because the Canada pension plan has this data to our to administer the CPP, they need to know when you died to stop sending out the checks. They need to what know your earnings history was, to pump it through their formula for what your benefit. So they need to have all this data. And under the Canada Pension Plan Act, it was determined there was room for researchers to access this data to help improve public policy around pensions. And so that was the way we were able to get access.

Cameron: Okay. Is there any other kind of qualitative information around that record identification that would allow you to do analysis on, say, gender or province or something like that?

Kevin: It turns out we know gender. We don't know province in our data. We had to make a case for each variable that we were able to access. And again it's only going to be information that's kept for CPP administration. So, you know, we don't know your education level or the name of your kid or anything like that, because that's not relevant for administration of the CPP. The CPP would know, for example, your mailing address, but that's not something that was needed for our research so we did not gain access to that. We did not access province. It would have been interesting to do some cross-provincial comparisons, but we did not access province in our database.

Cameron: Well, you've got enough on your plate with this because it's a fairly large database you ended up with. We'll look at that in a minute. I wanted to talk to you just a little bit about the structure of your paper, because I want people to understand how these papers are organized. So you start out: in your introduction is where you're trying to get the reader to understand the importance of the paper. And so there you're leading with the story you just gave us about the fact that there is this gap between rich and poor when it comes to longevity.

Kevin: Yeah, when you're writing an introduction to such a paper, you want to motivate that the question is important, and that can be important for the two reasons that I mentioned earlier. One is that the paper is going to address some area of theory where there is conflicting evidence, or there isn't evidence. That allows us to better understand theories of how the economy works. Or another motivation is that this is an important question for public policy and and that the answer to the questions posed by the paper will be useful in improving our public policies

Cameron: The next thing you do in the paper -- and this is pretty much unique to academic research that you don't see in something like journalism, I think -- is you've got this long section where you summarize everything that's relevant from previous studies, so that you can put your particular study in the context of what we already know. So in this case, what are the most important things that we didn't know already, that you can build upon in your paper?

Kevin: When looking at what we call sometimes the "literature review" section -- although I remember when one economist advising students said, "No one ever wants to read a section called literature review," so you try to find some other way to further that -- I just think of it as context for our paper. You try to do a couple things. One is to warm the reader up to what else has been done, so you know a bit of the history of the research in that area. And the other thing is to set out your case that your paper is a new contribution. In academic publishing, it's absolutely pivotal that you make a case, and a strong case, that your work is new and unique, and to use the football metaphor, that you moved the ball down the field. Really for most journals, that is the most critical part of the editor's decision. They say, "Yes, this paper looks right and I guess the question is interesting, but seems like, you know, you're not doing much different than anyone else has done before so, I'm going to reject the paper." Versus, if you can make a case that no one's done what you're doing before, and the results are really interesting, then that really increases the probability that you will get into a higher status journal and a higher visibility journal, and your work will get more eyeballs that way.

Cameron: Right. One of the things that I think you're accomplishing in your particular lit review in this paper, you and Tammy, is a methodological thing. So you're summarizing what previous studies have done, but you're also pointing out that they're doing it in a different way. They are more or less hand assembling a database by matching mortality records to unemployment records, or something like that, and they end up with a fairly small but useful database. And you're using a mass of data that allows you to do something different.

Kevin: Yes, that's right. Previous empirical work and data work in this area has often done one-off matches of a certain data set with some kind of mortality records, and part of the challenge of that, especially if it's a survey with only like 10,000 people or something, is that so few died in any given year that you don't have a very large proportion, you don't have a lot of mortality cases to deal with. I know this is a bit sombre a topic, but but there you go. Not enough people die to make it easy to do the work, which sounds like a horrible thing to say.

Cameron: No.

Kevin: The advantage of the data set that we have is that it was comprehensive, both in terms of coverage (it had all Canadians who paid into the CPP), and it also covered the entire span of the Canada Pension Plan from its beginnings in 1966 up to, I believe, we had 2015 or '16 at the endpoint.

Cameron: Right. When I'm teaching pension accounting to my students, I have to remind them that what constitutes good news for you is bad news for the pension plan, and vice versa. If you die, that's good news for the pension plan because they no longer have to pay the benefits.

Kevin: That's right. When you think about insurance, in the life insurance context, the bad event that you're insuring against is that you live too long, which is again a bit reversed from the way that normal humans think about it.

Cameron: So you end up with this fairly large database, compared to previous studies, and it allows you to do some different things. The one thing that I wasn't clear on -- because this is not really my area of expertise, this kind of data analysis -- what is the difference between a cohort analysis and cross-sectional analysis?

Kevin: For looking at mortality, when you see the release on Statistics Canada about life expectancy or mortality rates, the way they do that is that they look in any given year, and they look at how many people of each age died in a given year. So, like how many 50-year-olds die compared to how many 50 year-olds there are. How many 53-year-olds died, how many 53-year-olds there are. And they do that right from age 0 up to age 110 or so. And then they aggregate that across all those ages to get a measure of life expectancy. With the assumption being, what if today's 53-year-old behaves when they are 60, as today's 60-year-olds do? In terms of their mortality, behaviour. And there's an assumption in there, because today's 53-year-olds are not today's 60-year-olds. They are different. So when you do that with any given year, that's what we call a cross-sectional measure of mortality. You just, looking across who you see today, and assuming that behaviour of today's, you know, 50-year-olds is going to be the same as 50-year-olds in the future. No, the challenge with that, of course, is that different cohorts have different experiences. Some cohorts lived through the Great Depression. Some lived through deprivation in World War II. Some lived through the obesity epidemic of the past 20 years. Some lived through an era when smoking was normal, and some didn't. And so those cohort experiences carry through and affect life expectancy in a way that's not going to be captured in those one-year-at-a-time measures. And so, of course, the challenge with looking at what happens to a certain year of birth as they flow through their life, is that if you think about what happens to, you know, the change in smoking behaviour that started in the 1960s, well we kind of have to wait, what, 100 years, and for all the folks born in the 1960s to die off until we can actually answer that question using cohort methods. So that that requires a lot of patience by a researcher. So you can see why we're kind of anxious and would use the cross-sectional method to get an answer more quickly. But also it's perhaps biased because it doesn't necessarily reflect the behaviour and the experience any particular cohort. So what was attractive to us about the Canada Pension Plan data is, given that it has 50 years of history going back to 1966, we can see someone who takes their pension in 1973, and if it's now 2015, we can see what happens 42 years later: did that person who took their pension in 1973 at age 65, did they survive to age 93? We can look at that. And that's not going to be relying any assumptions about anything. We can just observe, for that exact set of people who were age 65 in 1973, how many of them survived to each age subsequent to that. And that is great from a researcher's point of view, to try to understand what was actually going on, because it doesn't require any extra assumptions about stability across years. It's just actual, raw data that we're able to use to calculate it. Of course, what you need to do, why researchers haven't done this in the past, is they didn't have access to this data and they'd have to wait 40 or 50 years for these results to out to roll in.

Cameron: Now, you mentioned in your paper that this approach allows you to compare Canadian and US data, and you've got this phrase about that. You're trying to see what may be driving the "steepening mortality gradients in the US." Is this what you are referring to previously about the discrepancy between rich and poor? Is that the gradient?

Kevin: Exactly. The gradient is just looking at people characterized by their earnings when they were working. So we put them in different pots, depending on whether they are a high earner, lower earner, middle earner, and then we look at those people and see how long they survive. And then we say, people who were low earners in their lifetime, how long did they survive versus people who were higher earners in their lifetime, and we see how they survive. In the US, there was a study published in 2016 in the Journal of American Medical Association, so a big prestigious medical journal, and it looked directly at that question for the US and found that people in the bottom 5% of their earnings while they were working, in comparison with those people in the top 5%, and found that there's a difference. So the gradient as you go from low- to high-earners was about 12 years in life expectancy, between low and high earners. So in Canada, we found that this difference is more on the order of eight years. So the bottom 5% versus the top 5%, the difference for men is about eight years of life expectancy. Which, you know, we think is a big number. It's one-third smaller than the US number, in terms of that difference, but it's still pretty big. It's still about 10% of your lifetime. So if you happen to be a low-earning male, you live 10% less than a high-earning male, and that means there's 10% you fewer summer sunrises and picnics with your grandkids and all the other good things that life delivers.

Cameron: I can see the motivation for you! Your database, then, you end up with a sample of the CPP database that covers 11 million people. For listeners who aren't familiar with the CPP, because we've got some international listeners, can you characterize, like, 11 million people in the CPP, how big is that compared to the entire CPP database?

Kevin: The Canada Pension Plan is a contributory public pension that was started in 1966. So employers and employees pay in, and then that money is used to pay pensions of retirees. It's not as comprehensive or large as Social Security in the US. We have some other pension programs to supplement it, but it is the core earnings-replacement pension provided through government in Canada. So it's definitely earnings-related. We only see you in our database if you worked. For those people who didn't work, maybe for older cohorts of women, for example, who might have been working in the home, we don't see them in this database. Or people who might've been seriously disabled, we wouldn't see them in this database, because it's only for those who worked, who pay into and then get money out of the Canada Pension Plan. It's fairly comprehensive. Overall population coverage would be very high. You're int our database if you even contributed once in your lifetime. So one part-time job, one anything, gets you in there. It's pretty rare that people didn't at least work a tiny bit in their life. So it is fairly comprehensive. That 11 million, the number of potential people in our database, we requested only people, I think it was years of birth 1916 through 1956. So it was just people born in those years. So over those years, we have 100% coverage of anyone who contributed at least once to the Canada Pension Plan.

Cameron: I see. Okay. So the fact that it's 11 million rather than 30 or 40 million is because of the years of birth that you are selecting?

Kevin: Exactly. Tor a study we wanted to have people born between those ages because that allowed us to look at the longevity pattern for the set of cohorts that we thought was most interesting.

Cameron: So for each of these people, you've got records that show you what they earned each year, if they received any pension benefits, what date the pension started, what day the pension stopped, assuming they died at that point. Was there anything else significant in there that we should know about?

Kevin: We also see disability status, which we haven't dug into yet, but that's for future projects. Because there is, as part of the Canada Pension Plan, a disability component. So if you've been working and become disabled, you can essentially take an early pension through the disability benefit. So we see that as well, which obviously is an important element in thinking about longevity, because folks who become disabled may have other health issues that arise that lead them to die, on average, more early. So that's something we plan to look at in more detail in future work. This first paper out of this project was looking at the lifetime-earnings-to-mortality relationship.

Cameron: I know from my own work in the field of disability that the disability benefit of the CPP is not all that easy to qualify for. There are some fairly routine denials on initial applications. It takes some persistence to get that. So, you know, understandably, the data you have on disability would be at least partially incomplete because some people who are disabled would just give up reapplying. So I guess what I'm saying is, you're gonna end up with this collection of data that will have some predictable problems with it and some that you find as you get in there. What do you, as a researcher, have to do to the data set in order to get it into a point where you can do a valid analysis. How do you clean it up or deal with missing data, that kind of thing?

Kevin: It takes a lot of work and fortunately, both Tammy and I, this is not our first rodeo. We knew that, getting into this, with law administrative data like this, that you have to look into things like making sure that data are consistent. To give an example of that, in both the benefits part of the database and the earnings part of the database, there was recorded your date of birth. And you don't want to assume, necessarily, that the date of birth is the same in both of those because perhaps -- and again, this goes back to 1966 -- there's an awful lot of paper and pencil then gets inputted into some big old 1971 IBM mainframe or something. There's a lot of scope for user error there that would lead to discrepancies. So we checked for these kinds of things to make sure things were consistent. We developed a protocol for how to deal with inconsistencies. We looked at people whose gender might change -- and of course, genders do change sometimes, that's a natural feature of reality -- but in terms of the database, this is only a more recent thing where the government recognizes such things -- but in terms of the database, we wanted to be careful of that to make sure that we were tagging people appropriately, that they weren't switching back and forth. So making sure that when we were merging people together, they were the same people and that we were recording their data appropriately. So that was one issue. Another issue was to make sure that the earnings records were complete and intact and and sensible, and so you do checks like you look at the average earnings of different people, and we compared to those to outside sources like the census or other databases, to make sure that things were in the same range. That's the kind of thing you want to check for, to make sure that you have what you think you have and that everything lines up with expectations. So that took us months. We hired a research assistant. The data were only available to us in a heavily secure data lab in ESDC in Gatineau, Québec. So we hired a research assistant to work for us in the summer of 2017. He was there from May through August digging through this data and doing all these checks to make sure that what we were doing made sense.

Cameron: So you didn't actually get to take the data out to your own computers and do the analysis?

Kevin: No, absolutely not. Because of the sensitivity of this data, it remained under highly secure protocols in ESDC. Although, as of 2018, we were able to move it to a secure facility funded by Statistics Canada here at UBC. So it is under lock and key. I can't take it out of there, but I can just walk across campus, which is an easier trip than flying to Ottawa.

Cameron: No kidding. Now, you started with 11 million people in the data set and after you've done all this cleaning and eliminating things that are not usable, you ended up with, if I read your data tables correctly, 3.7 million men and 2.7 million women. Did you start out with equal numbers of men and women?

Kevin: Yes, approximately. I mean women do live a bit longer than men and have a bit better mortality profile, so you'd think there'd be more women. But there's a couple of reasons why we ended up with fewer women in our final data set. One is that, in terms of people who never earned and never contributed to the CPP, there'd be a few more women in that category, just because of the gendered nature of working outside the home, over the 20th century. The second reason is that our goal was to, again, put people in these pots based on their lifetime earnings and we decided to characterize people's lifetime earnings by what they earned between the ages of 50 and 54. So if you didn't appear in the earnings record in 50-54, we cut you out of the data set. We did a lot of analysis to understand who this cuts out and how it might bias our results, and so that's in the appendix of the paper. But the basic idea is that if we're going to people in bins based on their earnings ages 50 to 54, if we don't see you in 50-54, we have to remove you from the data set. So that's why we ended up with many fewer women, because especially in the early part of the 20th century, a lot of these women were not working outside the home.

Cameron: Right, and then in addition, you have to work a certain number of years to qualify. So if you're in the data set, but you're only there for, say, three years, that's not enough to be included.

Kevin: Exactly. So this is a choice we made. We are dealing with this kind of stuff over such a long time horizon. Men are not as complicated because they've been always in the workforce and there's lots of interesting dynamics there, but it's challenging in understanding women because it's not just that you have perhaps trends in mortality over time, and trends of earnings among those who are working over time. We also have big, important trends across who is working and who's staying in the home and who's out in the workplace. And a lot of researchers approach this like, "Well, women are hard so we're just going to throw them out of the data set and just do this paper on men." And that's one option we could have pursued, but we think it's important, Tammy and I both, that we shouldn't stop studying women just because it might've been harder. So we pushed forward and kept our analysis parallel of men and women, but with really important caveats that comparing women born in, say, 1930 to the women born in 1950, that evolves really different set of women because of their experience in working outside the home and and how that might impact our data set. So, we put in those caveats and warnings and footnotes in the introduction and the conclusion, but we thought it was worthwhile to learn something about women, rather than toss them all out because they weren't an ideal comparison group for men.

Cameron: Hmm. So, it does present some challenges, and there are gender differences in the data, so that's going to drive a difference in what you end up being able to analyse.

Kevin: Yeah, for sure. And what was interesting about that is, again because of the patterns of work in the 20th century, what you have is a lot of the women who look like they're in a lower earnings pot, because they didn't have a lot of earnings themselves, often might be married to a man who was a higher earning pot. And so their family circumstance was perhaps better than was indicated just by looking at their earnings alone. And so what you found is the gradient of longevity with respect to their own earnings was much flatter for women, and we interpreted that as just because a woman's own earnings in the 20th century was less indicative of her overall family circumstances than a man's earnings. This was because -- again, in the 20th century -- men's earnings were the larger proportion of family income.

Cameron: So just to clarify, then -- because we are talking about this rather than showing it using PowerPoint slides and stuff -- what you found was that this gradient that we talked about earlier, between the life expectancy of poor men and, moving up through the income bands, up to the wealthiest men, the life expectancy increasing by a total of about eight years over the whole continuum of earnings, when you look at same curve for women, you don't see as much of a gradient. And your assumption or your hypothesis here is that it's because some of the women who are showing as being in a poor income bracket are actually in a family that might be well off. It's just their own personal income that is lower.

Kevin: That's very well described.

Cameron: Okay. I want to make sure that I'm following you, too, not just the listeners. So, you end up with this really nice database that you're able to use to estimate certain things. You talked about estimating lifetime earnings. So you are using this proxy of about a four or five year period to estimate what might be their lifetime earnings?

Kevin: That's right. We tried different windows -- we tried 10 years, five years, one year -- to get a sense of what someone's lifetime earnings might be. Now, we could have taken the whole 50 years of data we had available and characterized people's lifetime earning that way.

Cameron: That was my question, really. I didn't ask it clearly, but yeah, why didn't you just use all the data? Why do use this little window?

Kevin: The reason we didn't use all the data is that to have a full lifetime set of earnings from 1966 to 2015, that would allow us to study only a couple of years of birth: because, say, people born in 1950, they only started earning in, say, 1968. To take a better example, let's say the 1930 cohort, because the CPP only started in 1966, we missed ages 18 through 35 for them. So we'd have a very narrow set of cohorts, actually it would be about 1950 to 1960, that we could actually have almost all of their lifetime earnings. We wanted to broaden that out to cover more years of birth, and the way to do that was to shrink the window of earnings that were necessary to get into our database.

Cameron: Right. So you also end up with being able to do more of an apples-to-apples comparison, I think, because if you have some people whose lifetime earnings you calculate simply by adding up all their lifetime earnings, and you have other people where you've only got, you know, from 1966 on, and you have to use an estimate, it's better from a methodological point of view, it's better to use the same estimate for all the people, is that right?

Kevin: That's right, we wanted to use the same way when we're putting people in the different pots, the higher earnings pot and the lowering pot, wanted to make sure we used the same basis, the apples-to-apples basis, to do that. So that's why we made that choice. And you know there's a trade-off there, that we might like to use all the earnings that we see, but we analysed how much that decision mattered, and we showed results using different choices for a subset of our analysis, and so that is exactly why took us an entire summer of working with the data. And when you're doing empirical research, there is literally a thousand decisions like that you have make up: where you cut things off and how you measure things. And for each one of those decisions, you want to document what you did and you want to test whether the results are sensitive to that choice. That's what the best careful research does, and that's the standard we were trying to meet.

Cameron: I'm always interested, with academic research, how theory enters into what you're doing. And there's one thing that you're using here, a Gompertz projection, which I think embodies a little bit of theory about how the relationship should be between age and mortality. Could you explain that? Can you explain it in plain English?

Kevin: I'll try. There was a, I'll call him a social scientist, because I can't recall at the top of my head his history, named Gompertz who noticed a relationship between mortality and age, that it followed a very specific pattern. We refer to it as "log linear," which essentially means that it follows this very, very standard curved shape. And it was really tight. If you you knew, for example, the mortality rates by age, for ages 10, 13, 14, all the way up to 80 or something, and you fit it to this line, this Gompertz line, this log linear line, you could use that to project mortality rates for other ages, because it really fit this line very closely. And what's really interesting is how accurate this is. It's called Gompertz Law, because it seems like mortality rates do follow this very, very specific and tight pattern. So we were able to make use of that relationship to extend our data set a bit, because we don't see many people to age 100 in our data set. We would have to wait another 20 or 30 years to gather people to age 100 in our data set. We didn't want to wait that long. So we make use of the Gompertz Law to fit the data we see to this Gompertz pattern, and extend our data set a bit more to get projections of overall life expectancy that we don't see precisely in the data.

Cameron: So what did you learn from all of this? The next section of your paper is when you get into actually discussing the results, and you've got two sections there. You've got your main results and you got your extended results. I'm interested in the distinction between them. So what did you find, and how did you organize presenting those findings?

Kevin: The main result is to just see the gradient we have been discussing. The other main result is how that gradient has changed through time. So when we look at people born, say, in the 1920s and compare them to people born in the 1950s, did that gradient get steeper or more shallow or what? Again, in the United States, the experience has been that the changes across time have been very much in favour of those who are higher earners, and almost no life expectancy changes for lower earners. But what we found in Canada, which was strikingly different, is that while there been improvements over time in Canada, it's been almost uniform across different earnings groups. Lower earners, medium earners, and higher earners have all seen their longevity curves move about the same amount forward between the 1920s and 1950s birth cohorts. That's hugely interesting and hugely important. It's interesting because this is a fact about Canada that it's really important to know, that while there is an important distinction between being rich and poor in terms of your longevity, at least it's not getting worsen. I mean, that's something we want to know. It's different than the United States. I have seen people seeing the result in the United States, that mortality gains are only going to the rich, and trying to apply that to Canada. It turns out that you want to be careful in doing that. But then the second thing is, understanding different theories of why is it in the United States that this longevity change has been pro-rich, has helped higher earners more? And there's been a number of different theories about what might be driving that. Seeing what's happening Canada helps us know. It gives another example, another circumstance, to test some of these theories. If gives some evidence, not necessarily a test, but it gives some evidence that leans towards one theory and perhaps against some others.

Cameron: Can you go back to your point about the US? I've got a copy of the chart in front of me, Figure 3 from your paper. It shows quite dramatically that in the first quintile -- so the poorest earners, for people born in the 1930s versus people born in the 1960s -- for people born in the 1960s who are poor actually had a lower life expectancy. Is that what I'm seeing?

Kevin: In the US? Yes, that's right. It actually got worse. People in the bottom 20% of the population, the life expectancy was worse after age 50 than people born in the 1930s. That's an amazing fact. Essentially, the bottom half of the income distribution in the US saw no gains.

Cameron: And yet, on average in the US, life expectancy did go up.

Kevin: That's right, but it all went to the top half.

Cameron: Huh. I can see the importance of your study!

Kevin: Yeah, well, this matters. It matters for understanding our society and what's going on, in one way, but it also matters a lot in thinking about how to design pension policies that are fair, in understanding the role of our institutions like our public health system, of our other social supports, and trying to understand whether those things might be, in part, behind the fact that Canadian life expectancies have been improving at the lower end of the income distribution as well as the top. Why is that the case in Canada and not the US? Thinking about what social differences might drive that is really an important question, we think.

Cameron: The implications of this kind of study are pretty obvious for a pension plan. I mean, if you predict that the average worker's going to be living five years longer because of improvements to life expectancy and other factors, and then it turns out that it was only the richest employees that live longer, then your pension costs are going to be a lot higher, because the benefits are higher for them.

Kevin: Exactly. So if you were to talk to an actuary from, say, the Ontario Teachers Pension Plan, for example, they'll tell you that teachers in Ontario live longer than the average person. And they compare mortality rates of their beneficiaries to the average rates that came out of Stats Can's regular average mortality rates. They noted, and they know very well, that their population in the Ontario Teachers Pension Plan ends up being quite different. They knew that already, but what we were able to do is, you know, not just doing one pension plan at a time, but looking for the whole population in the Canada pension Plan to fill in that gap. Because the thing about people in the lower part of income is that very few of them are in these occupational pension plans. So we really didn't have great data on what their longevity profiles have been. And so that's one area where our paper contributes.

Cameron: Before we wrap up the discussion of the paper and move on to this other topic I want to talk about, about what happens next with your research once you've done it, was there anything else in that paper that particularly caught your eye that you were excited by?

Kevin: One last thing: in the US, two researchers Ann Case and Angus Deaton, revealed and uncovered -- "uncovered" is a better word -- they uncovered that mortality rates for middle-aged, white, non-Hispanic males -- so white, non-Hispanic males, a very particular population, and a big population in the US -- the mortality rates of that group were actually getting worse. And it's striking that when we see these longevity graphs across countries, they're always getting better. Right? We're thinking things are getting better, healthcare is getting better, our knowledge is getting better, our health is getting better. For this population, starting in about 2000, so 2000 to 2015, middle-aged, white, non-Hispanic males were actually dying more often than before. This was striking. I mean, it is different than any other country. Different than any other demographic within the US. This quickly became known as the "deaths of despair" literature, because they linked it directly to higher mortality rates caused by suicide, drugs and alcohol, and other (what they referred to as) deaths of despair. And this was linked into challenges in the labour market over the last 20 years, and how people in the bottom end of the income distribution haven't been doing as well -- you know, the decline of manufacturing in the Northeast, lots of things about these general, big, economic trends. It actually leads to an analysis that suggests that these things might be important in understanding why mortality rates are actually getting worse for this population. This was just shocking to all kinds of social scientists, that this was happening. And so, we looked for a similar kind of pattern in Canada and we found that we were not able to pick out white, non-Hispanic males, as we don't have racial data or ethnicity data in the Canada Pension Plan. We can see males. So we did pick out middle- and lower-earning males born in the 1950s. What we saw was, we didn't see a reversal in longevity, but we definitely saw a slowing down in the gains in longevity for that particular group. And so that suggests that this kind of -- we call it the Case-Deaton effect for the two American authors who uncovered this -- we were looking for evidence of this Case-Deaton effect and we found some indications of things we want to watch for in the future. We didn't see a mortality reversal, but we did see a slowdown in the improvement. And so when we think about the overall trends in inequality, thinking of what's happened to the lower end of the labour market and the people who are struggling a bit more in the labour market, some of those trends have been hitting Canada, as well. We want to keep our eyes open for these kinds of mortality shifts that might happening in Canada as well.

Cameron: So this leads really nicely to my next question, which is, when you've done a study like this, how does it end up actually affecting public policy? I mean, you've identified a couple of particular issues in the data set that tell us something about the way that our social policies are affecting Canadians, and there are some potential takeaways for things that could be done differently because of what you've learned. How does what you've learned actually translate into changes in public policy? Like what has to happen? What's the missing link between you publishing a paper and policy changing?

Kevin: Tammy and I thought about this a lot, about how we wanted to publish and present our work. We decided that the route we wanted to go was to produce a paper for an academic audience and also produce a paper for more of a policy audience. And we are both associated with the CD Howe Institute, and so at early stage of research we discussed this plan with some folks there, and we decided to do exactly that. We have a more academic version of the paper that digs deep into Gompertz projections and deep robustness checks, and we also have version that went out with the CD Howe Institute, which is much shorter and has more policy-based, easier-to-read focus to it. So that's how we designed our approach to influencing: making sure our work got some attention. We realized that thick, academic papers with 40-page appendices are something that a lot of very smart and interested people just don't have the time to dig through. So we wanted to present a version that was more accessible, as well. So that's how we did that.

Cameron: Can you describe what the CD Howe Institute is about?

Kevin: The CD Howe Institute is a think tank here in Canada. It publishes work in many areas. It has expertise in trade and tax policy and pension policy, and in health policy, along with other areas as well. The way I view think tanks is as a really important medium for doing exactly this, for walking between academic research that's published in dusty journals that many people don't see, and the real policy-relevance kinds of things that can lead to direct policy changes. I'm aware of several examples of policies that come out as a think tank working paper, be it CD Howe or somewhere else, that end up directly influencing and being adopted by different parties' platforms, or in their agendas when they form government. Think tanks have a vitally important role in translating academic research into policy-ready things. And they do that with their publications but also with events they convene. I am on the CD Howe Fiscal Policy Council. I know Tammy sits on their Pension Policy Council. What that just means is, a couple times a year we get together, they have people from the Department of Finance, from Ottawa, from Ontario, from some provinces, they have people from the Canada Pension Plan Investment Board, and also some private-sector insurers, and actuaries, and some academics like me -- and we get together and we talk about public policy issues and present some papers and talk about future directions for research. And I get a lot out of those interactions because I see people from industry and from the public sector are thinking about things. and I introduce them to some academic research they haven't seen, and they introduce me to questions that are on their policy frontier of what kind of things they need to know, and I can start my wheels turning to answer those questions.

Cameron: How does one as an academic get connected with CD Howe Institute. Is this something that you initiated and applied for? How does that relationship work?

Kevin: For me personally, it wasn't too hard at the time. My thesis advisor was also the president of the CD Howe Institute. So the connection wasn't hard to make.

Cameron: So nepotism!

Kevin: Yeah. More generally, whenever you don't have that fortunate circumstance, they always have their eye open for academics who care about Canadian public policy, who work in the areas of interest that they have, and they make connections. They sometimes ask me if I know of any more interesting people who are working in the area that we should invite to our conferences. And so the connections are made are often in that way. And I'm sure they would take unsolicited approaches, as well, and if they can find something that's a topic that they are interested, they would entertain offers as well.

Cameron: You also do a fair bit of work in the media. You write in the Globe and Mail and Maclean's magazine. How did you end up with those gigs?

Kevin: I write now and then for I those on the publications. I've been doing that since at least 2010. For the Globe, back in 2010 or '11, they started a feature they called "The Econo-something" -- I can't remember what it was now -- where they explicitly went out and sought a bunch of economists to write for their online publication as a regular feature. Maclean's then copied them at that, and their feature was called Econo Watch, and they also solicited some contributions from several of us. I started writing regularly for them and that led into some more of the print publications, as well. I became known to their editors and so they sometimes would approach me when a topic of intense public interest came up that they knew was in my area of research and expertise. They would approach me sometimes to write, and so that is how I kind of got into their world as well. But again, for those who work in these areas, they do take unsolicited contributions as well, and they do publish them. So people who are interested in getting into writing op-eds and those kinds of things, it is something that takes practice to learn how to write one of those things, but I know that you have an open window to consider submissions.

Cameron: I first ran into you on Twitter, through your social media account there, and I find the content that you put out is really clear to me, like, I really understand what it is you're saying. You have a real gift for communicating things in, well I guess it's 280 characters now, but 140 characters earlier. Do you enjoy the social media scene?

Kevin: Well, there are good and bad parts of it. You know, thanks for your kind words about what you've learned from them. I'm glad to hear you find it informative. In terms of any skill I might have in communicating, really it does come with practice and effort. It is not easy to learn how to communicate stuff in 280 characters. It is not easy to learn how to write an op-ed. It does take practice. And so it comes a bit with perhaps spending too much time on Twitter, too, but the good side of that is you can improve your skill at it. So the good side of it is that I do part of my role as an educator here. I'm paid the bucks that I'm paid here to educate people, and I view part of that as educating the public. The great magic and wonder of social media is that I don't need to be intermediated by any journalist, or any media, I just can communicate directly to people like you if you want to read the stuff that I put out there. That just really enhances my role as an educator. And I also try to take that to heart, that my Twitter feed tends to perhaps have a bit less snark than some other ones, and a bit less "what I ate for lunch," and a bit more, you know, "Oh, here's an item in the news and here's some background research papers," or "Here is a relevant government publication that allows you understand the issue." Because I really take my role on there as one of adding some value to the public conversation, and embracing that role as an educator through the social media account. That said, everyone has moments on social media. The danger of having that tweet button right in front of you is that occasionally you fall short of your expectations and you put out something a bit too snarky or a bit too whatever. And if you've got to pull those back, that's part of the experience of social media. So that's one negative side, is I've occasionally crossed lines that I regret, and I try to deal with those as as best I can. But also, the other side of social media is that it is a place where you get a lot of trolling and a lot of attacking. Me being a middle-aged white male, I don't get a lot of the same very personal attacks that many people of different backgrounds get, whether you're a person of colour or a woman or whatever. They seem to get it a lot harder than someone like me does.

Cameron: Mm-hmm. Well, Kevin you're an example to many of us in the academic world for the effort that you put into communicating your work beyond the academic journals. I really like the work you're doing with the Globe and Maclean's, and as I said, I enjoy your work on Twitter. I hope that there's some lessons that we can take away from this. It's hard to characterize these, because as you say, everybody has a different experience on Twitter, and the vulnerability of people who are in marginalized groups in our society and social media is pretty awful to watch. So, at any rate, I want to thank you for being on the podcast. It's really an education to listen to explain this stuff.

Kevin: Well thank you for giving me this opportunity, and I hope your listeners enjoy it, too.

Cameron: Super! Thanks, Kevin. Bye now.

Kevin: Thank you.

Links

Kevin Milligan’s bio page at UBC.

His research blog.

Kevin Milligan on Twitter.

Co-author Tammy Schirle’s bio page at Wilfred Laurier University.

Tammy Schirle on Twitter.

In the episode, we discuss the Gompertz projection. You can learn more about it here on Wikipedia.

The website of the Canada Pension Plan.

Credits

This episode was recorded using Zoom.

Host: Cameron Graham
Producer: Bertland Imai
Photos: Twitter and CBC (click photo to see article)
Music: Musicbed
Recorded: June 4, 2019
Location: York University