Thursday, August 13, 2009

Then Again, Maybe Not - Family Registry Version

After composing my post of yesterday, I could not get over the really odd idea that, if one goes by the family registries of the local administrative areas, the number of deaths in Japan soared 4.1% in between 2007 and 2008 but only grew 0.8% in between 2008 and 2009. I simply could not accept such a swing between two consecutive years as reflecting anything even approximating reality.

So I looked back in time to see if there were any other anomalies.


The columns are the number of deaths recorded in family registries, according to the annual survey conducted by the Ministry of Internal Affairs and Communications. The red line is the year by year growth in the number of deaths, in percent.

Hmmmm...

Uhhh...

Hmmm...

All I can say is whatever it is that the family registeries statistics are registering -- it sure ain't likely to be the actual number of deaths in any given year.

3 comments:

Jan Moren said...

I'm not so sure. You're looking at a first derivative after all (the rate of change) and it's inherently rather noisy (second, third and so on are progressively worse, which is one reason systems control, in economy as in mechatronics, is really hard). Those swings - a couple of percent increase or decrease on a yearly basis - do not look terribly strange to me.

At the minimum you should probably see if you can correlate the swings to the size of individual cohorts (a larger cohort entering a stage of rapid decline would shoot up the rate a bit, for instance), and to known common mortality factors among elderly such as the severity of the prevailing influenza epidemic each year. In fact, I would not be surprised if something like the seasonal flu couldn't explain most of that movement.

MTC said...

Herr Moren -

I too thought about increased morbidity due to influenza as a possible cause of variability. Given the size of the population, a bad influenza season would have to result in an additional 40-50,000 deaths annually to account for the swings in the registered deaths. That seems like a lot. I would have to look at the deaths attributed to pneumonia in the Health, Labour and Welfare statistics to feel any degree of comfort with such an explanation.

Zach Baran said...

"That seems like a lot."

Can I venture to say that the whole argument in question has emerged from this cognitive bias? My presupposition is quite the opposite: I don't see a reason why the numbers of births and deaths tallied from family registers would be wrong. If there is one statistic that a government needs to get right, it would be the number of citizens that appear and disappear each year.

As for what it seems to say, you've already shown that similar swings have been recorded in the past. What are the drivers for these swings? I don't know. Looking at the population statistics available from MIC (which conveniently start from the year following the 'hinoue-uma' downward spike in the population), there are large swings in both births and deaths.

Population data is not seasonal, and so I don't really expect smoothness in the data when slicing a long series of ebbs and flows from 1978 - 2008 into thirty equal parts. For instance, I could look at that spike in 1998 and assume that, because I don't have in front of me an explanatory variable for an increase of over 150 dead Japanese per day compared to the year before, the government must have gaffed the statistics. Then again, if I start hunting around and find that there was also an increase of 23 suicides per day compared to the year before (and quite a spike, historically), I'm on my way to explaining the variation, as getting over my cognitive bias.