Or dental floss to clean your teeth after chewing on the Case-Deaton study, discussed in my earlier post. Andrew Gelman makes an interesting point about what might drive some small part of the findings (that even though middle-aged people in lots of other countries and among US Hispanics and non-Hispanic blacks still enjoy declining mortality rates, it seems that non-Hispanic white Americans in that age group do not.).
That point is this: The age group that Case and Deaton studied, the ages from 45 to 54, has not had a constant average age over time.:
But could this pattern be an artifact of the coarseness of the age category? A commenter here raised this possibility a couple days ago, pointing out that, during the period shown in the above graph (1989 to the present), the 45-54 bin has been getting older as the baby boom has been moving through. So you’d expect an increasing death rate in this window, just from the increase in average age.
If you like, the load in the boat containing all 45-54 year-old Americans has been tilting towards the higher end of that span, because of the baby boom effect.
Gelman does some back-of-the-envelope calculations and suggests that the correct death rates for middle-aged non-Hispanic white Americans might not have increased, after all, but stayed constant. This is still different than the evidence from other countries or the evidence on American Hispanics and non-Hispanic blacks, because all those groups seem to have enjoyed declining mortality rates.
The question of interest for me has to do with non-Hispanic blacks. I get that American Hispanics might not show the same baby boom effect if their countries of origin didn't demonstrate a baby boom. But many European countries (though not Sweden) in this graph should show similar effects to the tilting of the boat of the middle-aged towards higher ages:
The group I'm most curious about in this context is the group of non-Hispanic black Americans. Didn't they experience the same baby boom effect as non-Hispanic white Americans? Studying that question could throw more light on the findings.
Why am I harping so much about these details?
When we find something very unexpected and shocking, such as the sudden increasing mortality rates of poor white women (but not of poor black women, say) or poorer non-Hispanic whites between the ages of 45 and 54 (but not of poorer Hispanics or non-Hispanic blacks), we should double- and triple-check all the calculations. That's because it's hard to make up explanations which would explain those ethnic and/or racial differences. For instance, poverty-based explanations shouldn't work differently on whites, blacks and Hispanics.
Or put in another way, before we launch all the necessary extra studies about these phenomena we should be more certain that they are real.
One simpler way to describe
Most of these studies are about changes in a ratio, one where mortality is the numerator and where some population measure is the denominator. It's natural to interpret changes in the ratio as coming from changes in the numerator (such as increasing death rates for poor white women), and that can be the case (and perhaps is). But before we conclude that, we should make sure that the denominator hasn't changed.
By that change I don't mean quantitative changes* but more the idea that who it is we are counting in the denominator may have changed. Gelman's point above is an example of that type of a change. Another similar example is mentioned in this older post of mine, about the rising mortality rates of poorer white American men and especially women:**
Suppose that the group "white people without a high school diploma" has shrunk not only in proportion to the overall population but in proportion to all whites. If that's the case, it could be that past studies of similarly defined groups had more people with higher life expectancies in them, but that the most recent group does not, perhaps because education has become more accessible, filtering away first those with minimal risk factors?
That one is about who it is who remains in the "least educated" groups over time, and this could differ between white, black and Hispanic Americans.
So what can we conclude about the Case-Deaton study? Certainly that the numbers deserve more investigation.
Edited later to add first footnote
* That explanation is a simplistic one, for which my apologies. The changes in what types of people are included in a certain category (by age, gender, race, geographic area) obviously can affect both the numerator and the denominator. But my tool should work as a pedagogical one, to remind us that when a ratio changes it's not necessarily (or only) the numerator that is changing.
**One newer study, still in a draft form, suggests that changes in the distribution of education might not explain the rising mortality rates of poorer white women, but might do it for poorer white men.