Wednesday, August 19, 2009

Variability in Recorded Deaths - A Response to Comments

In comments to my post of August 13, Janne Morén and Zach Baran cautioned that I may be overreacting to what looks like a peculiar variability in Japan's annual increases in deaths recorded in the family registries. The year-on-year swings in the increase/decrease seemed quite large and without an obvious justification.


For the sake of comparison, here are the graphs of the same data for the United States and, in Herr Morén's honor, Sweden, over the same time span as the above graph for Japan.

First, the United StatesSources: U.S. Census Bureau, CDC

Now Sweden
Source: Statistics Sweden

Looking at the data, there is the obvious difference that the number of persons dying in Japan is increasing at a steady pace -- with bizarre accelerations and decelerations along the way--while the annual number of deaths in the United States is growing only moderately and Sweden's numbers are falling. Looking at the moves in the Japanese data--the first derivative--in comparison with the shifts in the same data for the U.S. and Sweden, I still feel a sense of extraordinary unease at number of jumps of over 3% year-to-year.

Anyone else care to offer his or her opinion about what, if anything, is going on?

2 comments:

Jan Moren said...

Again, with the exception of that 6% spike (wonder what happened that year) I don't really see anything odd. The overall variation seems to be reasonably similar. A careful statistical analysis might uncover something, though you'd had to control for the differing age ranges of the deceased (a younger overall population would be less affected by typical old-people's issues like cold winters, heat waves and the flu). That would be rather difficult to do with such limited data, though.

The one thing that hits me is that given the current trend in Sweden as shown in the last graph, and given that the population is not decreasing but increasing, which would mean that the expected average life span increase in Sweden exponential and unbounded at the moment. If the trend continues the whole nation will eventually be effectively immortal.

Or, this kind of short term data has too much variability and noise to be of much use for any detailed examination of trends and effects. Immortal cheap furniture designers or shaky data - the interpretation is up to you.

Zach Baran said...

I hate to be wonky, but we are not dealing with the first derivative. We are simply calculating the percentage change "year on year".

On a comparative basis, Japan's data may indeed cause unease. Beyond the parallel spike in suicides (especially among males aged 50+), I cannot find other factors that would explain the increase.

The Ministry of Health, Labor, and Welfare publishes statistics on the five leading causes of death, and during the mid-1990's the top five changed to (1) cancer, (2) heart disease, (3) cerebrovascular diseases, (4) pneumonia, and (5) accidents. This order is unchanged as of the last data I saw, which was from 2007.

Here's the rub: All five leading causes increased from 1997 thru to 1999, but the increase were much more dramatic between 1998 and 1999. However, the total number of deaths in 1999 was less than the number from 1998.

So is there so much variation in the more minor causes of death (traffic accidents, poisoned curry, etc.) that could account for the spike in '98? I don't know, but if the data used by Dr. Okamoto at this site is to be believed, than there just may be:

http://vetweb.agri.kagoshima-u.ac.jp/vetpub/dr_okamoto/Moromoro/ST4.htm

I suspect that changes in the way data was reported account for much of the volatility seen in "causes of death" (e.g. heart disease), but for something slightly more easily discernible like births and deaths, that Japan has been compiling for over 1300 years, I want to use as a base assumption that the total figures are correct.