Ilya Kashnitsky

This post is based on my previous Twitter thread.

Demography 101


and (even more important)

โŒ what it isn't โŒ

Join in for the most topical demography primer

๐Ÿงต THREAD 1/x

— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Unlike many statistics and quantities of general use that we tend to see regularly, life expectancy is not observed directly. It's an output of a *mathematical model* called life table.

So, why can't we do without a model?


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Consider a seemingly simple task: you want to know how long people live. What can be easier? Let's just see how many years lived those who died recently.

Why not?

Such a metric would be massively driven by population age structure


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

For the most of the recent history human populations were rapidly growing, which means that each next generation was bigger than the previous one. Relative differences in the size of generations affect the age composition of those dying.


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Okay. Then why don't we simply take a group of people born in the same year (demographers call such groups cohorts) and see how long on average they live?

We could. But it takes remarkably long to wait until the last one dies. And we want to know what's happening *now*


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Okay okay you irritating Dr Limitation. How can we learn what's happening now?

Well, for that we need a mathematical model. But as any model it comes with certain assumptions and limitations*

I take them, tell me!

*that we should not forget once the precious results appear


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

So, period life table

(as one can guess there's also a cohort life table, but for the data requirements outlined above it's not an option to explore what's happening with mortality now)

The idea is simple: take those dying now and divide them by the size of their age groups


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

This yields age-specific death rates โ€“ the key input for the life table needed to calculate life expectancy

Now, let's take an imaginary cohort and see how long would they live on average if they experience these observed age-specific death rates


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

The imaginary population is know as a synthetic cohort. And here comes the main assumption of the life table:

The model assumes that the observed age-specific death rates stay *unchanged* throughout the hypothetical lives of the hypothetical people in the synthetic cohort


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

This big assumption almost never holds in real life!

Mortality in human populations keeps improving beyond the most optimistic expectations. For decades the best demographers were systematically underestimating the progress in mortality reduction ๐Ÿ‘‡


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

If there is just one take-home message from this thread let it be

โœ… Life expectancy is a snapshot of the *current* mortality

โŒ It's not a projection/forecast of the actual experience of the newborn cohorts


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

So why do we talk about "expectancy"?

According to @therealrchung ๐Ÿ‘‡ this originates in the "expected value" meaning that came from statistics

A very unfortunate name for the concept that became so crucial in public discussions on human development


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Let me re-iterate:

๐Ÿ‘‰ Life expectancy is a summary measure of the *current* mortality ๐Ÿ‘ˆ

Here is a brilliant analogy by Robert Chung elaborating on the *current* nature of period life expectancy


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

The most popular error in public perception of period life expectancy forgets about the heavy assumption of the synthetic cohort
(constant age-specific death rates throughout their hypothetic lives anchored in current year)
and talks about the future of kids being born now


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

In normal years this large interpretation error is somewhat masked by the gradual and often close to linear improvements in mortality. A rule of thumb is to simply add ~6 years to period life expectancy to obtain a reasonable cohort estimate ๐Ÿ‘‡


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Mortality shocks like 2020 are a different story though. Here the "forward looking" (mis)interpretation of period life expectancy projects the *shock levels of mortality* into the future. Of course this doesn't happen. Shocks are called shocks because they are temporal


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

A notable example of this false reasoning went far and wide last week published in @statnews. Departing from the wrong interpretation of life expectancy, the op-ed estimated c19 years of life lost per person ๐Ÿคฆโ€โ™‚๏ธ

A rabbit hole RT ๐Ÿ‡ for you to explore


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Demographers tried to explain and mocked this piece a lot ๐Ÿ‘‡, but I'm afraid the harm is done and we are going to hear "pfff just five days" for many months to come

The seemingly easy life expectancy interpretation trap clearly demonstrated


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Another detail that often misses public attention is that life expectancy is not a single value โ€“ it can be estimated for every age

Most often and by default life expectancy is reported "at birth". But we can estimate remaining life expectancy for various ages


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

And here comes another popular misunderstanding of life expectancy. I bet each of you has heard something like this ๐Ÿ‘‡ at least several times in your life


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

True, there were times when life expectancy at birth was about 30 years even in the most developed now countries. This doesn't mean though that those who outlived this threshold age were getting old at young (by our current standards) ages

Let me illustrate


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Let's take Italian male population in 1872, the first available year in @HMDatabase. Have a look at the survival of this synthetic cohort โ€“ the proportion of the initial cohort that is still alive by certain age

Half of the synthetic cohort died by age 15!


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

And here is how remaining period life expectancy looked by age

Infant and child mortality was sooo high that those escaping early deaths had higher remaining life expectancy

At age 34 remaining life expectancy was the same as at birth

Only, it applied to the 41% survivors


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

And I guess the perception of age was not radically different among those survivors. It was all about selection and luck getting there

BTW, this links to another popular demographic myth of everybody having lots of kids in the past. No, people used to have lots of births


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

On this sad note I will wrap up my life expectancy primer. Feeling relieved to let out the idea that was occupying the back of my mind for months now

Next, I challenge @jm_aburto ๐Ÿ˜‰ to discuss lifespan inequality


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

And a couple of bonus links and RTs for those interested in diving deeper

Have a look at the beautiful explanation by @CSchmert in reply to the discussion initiated by @VictimOfMaths


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

A very deep and interesting discussion of life expectancy was recently published by demographers from Vienna and brought to me by @MarkusSauerberg ๐Ÿ™


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Fresh out from @PNASNews โ€“ย a paper by @jwvaupel @VillavicencioFG and @bergeron_mp that outlines the exciting story of human mortality development


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Should you wish to read a demography book to learn the nuances consider getting a copy of our discipline's Bible โ€“ Demography by Sam Preston, Patrick Heuveline, and Michel Guillot


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021

Let me finish with a RT of our most recent work that estimates life expectancy drops in 2020. I hope this thread would add to a better understanding of these results.

Open to questions and discussion ๐Ÿ‘


— Ilya Kashnitsky (@ikashnitsky) March 5, 2021