Mortality Nuggets: "Young People", Actuarial Club Meeting, and Disparate Results Pre-Pandemic

Some new, but a lot of old results

Last Tuesday, May 16, I was at an actuarial meeting: the spring 2023 meeting of the Actuaries Club of Hartford and Springfield. More on that in a bit.

But first, something old from the Wall Street Journal.

Yes, “Young Americans” Have Had a Bad Mortality Trend from Non-COVID Causes

WSJ: Young Americans Are Dying at Alarming Rates, Reversing Years of Progress

For decades, advances in healthcare and safety steadily drove down death rates among American children. In an alarming reversal, rates have now risen to the highest level in nearly 15 years, particularly driven by homicides, drug overdoses, car accidents and suicides.

The uptick among younger Americans accelerated in 2020. Though Covid-19 itself wasn’t a major cause of death for young people, researchers say social disruption caused by the pandemic exacerbated public-health problems, including worsening anxiety and depression. Greater access to firearms, dangerous driving and more lethal narcotics also helped push up death rates.

With this graph:

And this:

I assume the 2021 data were “provisional” due to academic publication issues. Because the 2021 data have been finalized for a long time.

“Poisoning” — that is going to be primarily drug overdoses. Except for ages 1-4, when it really is poisoning like you think.

Which really points out the problem of this analysis. It shouldn’t be lumping all “young Americans” together, because this is absurd. For mortality trends, there is a reason I split it up 1-4 (preschool or little kids), 5-12 (school age or just plain old “kids”), and 13-17 (teenagers). I don’t include 18-19 with these groups — 18-24 can be its own group of young adults.

I did a twitter thread starting here (I will embed it on old STUMP), but let me make it simple: it’s not all “young Americans” with a notable increase in deaths during the pandemic.

It’s teenagers.

I will write more about it later, but I’ve redone one of the graphs from that post so you can see:

That has all the causes from the stacked column graph converted into a line graph.

There is a lot going on, so let me just graph a few key causes of death:

Just focusing on the major causes that are influencing what’s going on, two of them had major trends pre-pandemic:

• Motor vehicle accident death rates had a fabulous decreasing trend from 1999-2013, after which it was a bit sideways

• Suicide death rates had been steadily increasing 2007-2016 and had a peak 2017-2018. Rates came down in 2019, and somewhat increased in 2020-2021, but were still below the 2017-2018 peak.

The other two had ambiguous pre-pandemic trends and definitely had huge increases during the pandemic.

More on that later. Lumping in the kids with the teens makes no sense, and they know it doesn’t, as they deliberately exclude infants:

Given the nature of little kid (age 1-4) deaths, school kid (age 5-12) deaths, and teen deaths differing, it really makes no sense trying to lump them all together. The magnitude and nature have very different trends.

Actuarial Meeting Notes: Mortality Differences

As my intro stated, last Tuesday I was at an actuarial conference, and I presented on group and individual life insurance. There is a section on mortality, if you’re interested. (It’s just the slides in PDF, not a recording of my presentation.) Go there if you want to see what one of my Powerpoint presentations usually looks like for work.

However, I do want to point out a different presentation on mortality experience, from some actuaries at Oliver Wyman, Lisa Grieco and Mark Spong. Titled “Mortality Improvement Trends by Socioeconomic Drivers”, it was a presentation of some of their work published at the Society of Actuaries last October… and I was intrigued by their results.

They got to do something I’ve looked at, which is to get super-special access to Census, etc., data (and this requires all sorts of clearance, yadda yadda) so that you can do research on disaggregated data - that is, use individual-level data.

But it’s a real pain in the ass… and I don’t have time/patience for this. So kudos that they did this!

They were trying to develop a model and test out how strong different variables were in explaining mortality differences:

This is a really geeking-out sort of thing, and the data itself has some interesting patterns (and I’m about to write something for work using this). There are a few patterns that make one wonder why certain divergences in mortality improvement (all of this is only up to 2015, so no pandemic experience is involved):

I will talk shop for a short moment — life insurance is very competitive in the U.S. along a lot of dimensions, and when actuaries develop our mortality assumptions, we don’t assume that mortality in the future will be the same as it is today. But we know that the “shape” of mortality improvement is not level — the Society of Actuaries has experience studies, across multiple insurance companies, and we also look at the general population.

We also know that different companies may have different mixes of policyholders — I used to work at TIAA, whose main client base is university professors. That’s different from Knights of Columbus, a Catholic fraternal group, or USAA, which is for members of the U.S. Armed Forces (active or retired) and qualified family members. These all have different customer profiles. Insurers will develop how they assume their business’s mortality will go into the future based on who their customers are, and the underwriting information they have.

So every so often a life insurer may have “bad mortality experience”. It doesn’t mean that more policyholders die than the general population for the same ages — it means more people died than the insurer had originally expected. And if the insurer expected the older folks to have had mortality to improve better than it actually did….

….well, this actually happened to many insurers before the pandemic.

And, it seems, something was happening in the general population where older people didn’t have as rapid mortality improvement as younger people, and, in particular, their mortality was getting worse (that’s what the negative numbers mean).

Back to the researchers - their research report is here: Mortality Improvement Trends, Independent Analysis on Socioeconomic and Other Drivers, and they’ve got a spreadsheet model I’m going to play with. Looks like fun.