More On That Most Important Study

Views

I wanted to follow up on my post from yesterday about that study on middle aged white mortality rates. The study generated a great deal of attention when it was first released earlier this fall. And it also generated a small methodological controversy. Andrew Gelman, a Professor of Statistics at Columbia University, was one of the principal critics and published a set of revised numbers that control for the size of population groups passing through the 45-54 year age bracket over the course the years between 1998 and 2014. Before digging into the numbers, I think the overall gist is that Gelman’s findings/revisions do not challenge the overall findings of the study – a conclusion Gelman himself seems to agree with. But they change the details significantly.

Let me first summarize the technical issue and differing results as simply as I can. Gelman noted that Case-Deaton had not controlled for the changing sizes of the age groups over the period in question. This turns out to be significant, not just a technical matter, since we’re talking about the period in which the baby boomers are moving through middle age into their 60s. To put this more concretely, the people who are 50 in 2000 are not the same people who are 50 in 2005. (That’s obvious). Less obvious, the number of people who are 50 in 2000 isn’t the same as the number who are 50 in 2005 – particularly if we’re talking about a historic bulge in the generational population.

Now, the methodological differences here are beyond my ability to independently evaluate. But having read both studies and read the discussions emanating from them, I believe Gelman is right on this point and I’m going to assume for the purposes of this post that he is correct.

So what difference does this weighting for age make? Pretty significant ones it turns out. When he recalculated the numbers, Gelman found that rather than a continuously rising mortality rate for middle-aged white people from 1998 to 2014, the rate rose as Case-Deaton found until about 2005 and then basically leveled off at the higher rate.

But things get even more surprising when he broke the numbers apart by gender. The leveling off conceals a starker reality. Starting in 2005, the mortality rate for men began to go down and got back to 1998 levels by 2014. The mortality rate for women kept right on going in the wrong direction.

So here’s the uncorrected, Case-Deaton trend line.

And here’s Gelman’s trendline, with the age weighting and gender breakdown.

As I said above, I think my analysis from yesterday applies just as much to these numbers as to the original ones. But the details are quite different. Gelman also notes that the trend is much more pronounced in the South and least in evidence in the Northeast. Again, for a phenomenon seemingly tied to race (and in my interpretation, shifting or collapsing racial hierarchies) that regional dimension is not surprising.

One other point is important to make with these graphs. The Y-Axis representing morality rates is not the baseline in this case. Basically every other group everywhere sees a constantly declining mortality rate. So flat is not good or normal. Indeed, with the chart above showing the gender breakdown, on first blush it looks like white men came close to making up the lost ground between 2005 and 2014. But that’s not the case. Middle aged white men have only gotten back to where they were in 1998 while almost every other group has made significant progress during that period.

To get a sense of the difference, here’s Gelman’s charts look at men and women, white, black and hispanic in three age groups over the period in question. (red=women, blue=men)

As you can see, to paraphrase Sesame Street, two of these charts are not like the others. Everyone else is on what looks something like a playground slide, albeit a somewhat bumpy one. Whites in the 20 year bracket between 35 and 54 are either flat or going up. And that applies to both genders and notably almost as much for the earlier age group as the first identified.

Why men and women seem to diverge so clearly after 2005, I have no real theory. There are plenty of possible explanations for why economic factors might affect women more severely – plenty of explanations for the opposite, too. But why this should show up uniquely for white women, as opposed to women generally, I’m at a loss to explain that. But again, big picture, two of these charts are not like the others. The global reality remains the same.

LIKE US ON FACEBOOK