February 25, 2022 | News | Interview

“The Biggest Change Has Come from the Data Itself”

In this Interview, Alyson van Raalte, Head of the Independent Research Group: Lifespan Inequalities gives an overview on how demographic research on mortality patterns has developed over the past 25 years, identifies two major challenges for the field in the future and explains why she thinks data availability also comes with a price.

Dr. van Raalte, why are demographers looking at patterns of mortality, more precisely, population-level age patterns of mortality?

The key task of demographers is to describe and explain changes in the size and structure of the populations. These changes come from changing mortality levels, but also from changes in fertility and migration. Those of us like myself who study mortality are asking questions like what ages and causes-of-death are contributing the most to life expectancy changes over time and to differences between populations. How and why does the age pattern of mortality differ across social groups? What are the causes of accelerations and decelerations in mortality decline over time, and what do we expect the future to hold? Lots of other social science disciplines are asking similar questions, but often their interest is at the individual level, e.g. if I changed something in my life, how much longer could I live? But increasingly, disciplinary boundaries are getting blurry, and there is a lot of cross-disciplinary work happening which strengthens all of the social and health sciences.

How have technological changes over the past 25 years influenced demographic research?

Like all of the social sciences, demography has benefitted from enormous technological change. The internet has eased the process of international collaboration, data sets are digitized and more widely disseminated, and software is more powerful. Probably the biggest change has come from the data itself; we make increasing use of longer-term datasets from population registers and long-running surveys that allow us to observe the same individuals repeatedly over their life course. In the past, we mostly worked with aggregated tabulations of census data or with population surveys catching a snapshot of the population at a certain time.

Why do you think data availability also comes with a price?

With more data covering all aspects of the life course, we are able to answer hundreds of increasingly detailed questions, which on the whole is a good thing. But this can also cause us to lose sight of the big picture. And even when we keep abreast of international developments in mortality, it becomes incredibly challenging to take stock of all of these micro-findings and figure out which determinants are actually making a difference to why life expectancy is higher or lower in one country compared to another.

In your recent review in ‘Population Studies’, you say that what we are learning about mortality risk is not always being adequately translated into its population-level impacts. Would you explain this thought and give an example?

One example I gave was the impact of educational expansion on mortality patterns. We have many excellent studies that have looked at trends in life expectancy by level of education, almost all of which show that the more educated live longer on average than those with lower levels of education. At the same time, younger generations are generally more highly educated than their parents’ generations. In some countries these changes in the educational composition have been enormous, and happened within just a few decades.

What kind of research questions should be asked to fill this gap?

Beginning with Evelyn Kitagawa in 1955, demographers have a long history of trying to separate out these within-group changes in mortality risk from compositional-changes in the population. But the biggest challenge is to account for contextual changes in the population. As the educational level increases, this might confer benefits to everyone in the population regardless of educational level. But at the same time, when attaining a very low level of education becomes a rare event, this group will contain a higher proportion of individuals who are unable to reach higher levels of education because of sickness—in other words, it’s not the lower education itself causing a higher risk of dying, but the differences in health which cause the differences in completed education—what we call reverse causality.

So, there are no easy answers here…

Yes, but if we are to keep our eye on the big picture, we need to think more about the role of compositional and contextual differences on population health, and not just focus on the causal impact of the determinants themselves.

What are the challenges for demographers in the future?

There are many. Let me explain two: First, the basic data infrastructure needs to improve. The Covid-19 pandemic has really laid bare to the rest of the world just how important it is to have functioning systems of civil registration and vital statistics. If we have bad estimates of the total numbers of deaths before the pandemic, there is no good way to make an estimate of deaths caused by the pandemic. And of course, this goes beyond the importance of tracking Covid-19 mortality; any evidence-based policy designed to improve health and well-being requires a solid data infrastructure to be evaluated. We need to register every birth and death around the world.

…and the second challenge you see?

We need to think of ways to integrate our theories of mortality decline into our descriptions of mortality change. When asked to explain why Japan and South Korea are leading life expectancy rankings, and the US is falling further behind, we really don’t have a simple answer. We can offer up a number of explanations, for instance differences in health behaviors, health care, the built environment, different selection processes from poor mortality conditions experienced during childhood, etc. Likely these are all contributing to some degree. But we really should be working toward better ways of testing these theories against each other so that we can actually rank some of these determinants and understand under which circumstances they are most important.

Background Info and Definitions

What is life expectancy?

Life expectancy is the hypothetical average age at death of a population where everyone experienced the current age-specific death rates for their entire life. In other words, it isn’t a projection of how long anyone will actually live, what we might expect from the word ‘expectancy’, but it’s a pure hypothetical.

We use life expectancy instead of comparing the actual observed ages at death across populations to control differences in the age structure of populations. If not, we wouldn’t be able to answer whether the average is different because mortality is lower at every age, or because one population has a younger or older age structure.

What do life tables have to do with it?

Life tables are the tool that we use to make our life expectancy calculations. Basically, we work through what the age pattern of mortality would look like if we subject a hypothetical population to the observed mortality rates at each age.

How has life expectancy developed in the past 25 years?

On the whole, there has been steady progress. The United Nations estimates that world life expectancy has risen from around 65 years, 25 years ago, to around 73 years today. All major regions of the world have a higher life expectancy today than they did 25 years ago, and progress has generally been faster in low- and medium-income countries (LMICs), leading to life expectancy convergence at the global level. This is great news.

But the progress has not been even. The HIV/AIDS epidemic in Sub-Saharan Africa, violence and war in several countries, and the transition to a market economy across many populations of Central and Eastern Europe all led to short-term mortality increases. And now of course, many countries have experienced life expectancy declines of a magnitude not seen since the second world war as a result of the Covid-19 pandemic.