Tuesday, August 11, 2009

More on fertility and the HDI

James Holland Jones, a biodemographer at Stanford has posted a critique of Edward's critique of the recent paper by Kohler, Myrskylä and Billari et al. in Nature arguing that development beyond a certain level reverses the fertility decline usually associated with development. Jones' response is a good and fair one and well worth reading, because it gives a good overview of some of the issues involved. However, much of the response amounts to "you work with the data you have, not the data you'd like to have". This is true, but it troubles me that Myrskylä et al. are making a fairly strong claim based on fairly weak data. I would like to add some of my own critique of the Nature paper, which I think complements the points Edward raised.

First, maybe I should say something about why Edward and I are devoting so much attention to this article. If the claims are correct, it really would have very important consequences. Almost everybody would agree that excessive aging is an important impediment to economic growth, both because of the sheer fiscal cost of paying for a large population of retirees and because of structural issues associated with aging, like increasing export dependency, increased savings, reduced consumption and lackluster growth. If the article is correct, this would imply a couple of things: first, that to a certain extent (but not completely) this is a problem that resolves, or at last alleviates itself somewhat as long as one can attain a certain level of development. Second, it suggests a different kind of low-fertility trap where countries who pass a certain tipping point in development can raise their birth rates whereas countries who fail to get to that point may enter a vicious economic circle.

This may in fact be the case, but I dont think Myrskylä et al. make a convincing case for it. Let's have a look at their graph:

Now what's important here, as Jones, Hugh and I all agree on is the black line, which shows adjusted total fertility rate that is not distorted by birth postponement. Looking at, it does indeed seem like there's a positive correlation between adjusted total fertility rate and the Human Development Index. However, in this case correlation does not mean causation. If you look at the countries with an HDI above 0.9, you can distinguish a few important groups. The Nordic countries ( 5 countries, average aTFR in 2005: 1.99 range: 1.92-2.2), English-speaking countries (4 countries, average aTFR: 2.01, range:1.91-2.2 Canada and New Zealand are not included in the data set, so UK, US, AUS and Ireland). These are two cultural groups that both have a great deal of gender equality, the Nordic countries also have a similar social model. There's a bit more variation in the Anglophone group in terms of economic model, but also many similarities. Together, these two groups dominate the upper range of the HDI and they all also have high fertility. Since Canada is not in the data set, they have the top five spots in terms of the HDI.

On the other hand, you have groups countries with more conservative family models: German-speaking countries (3 countries, average atfr 1.58, range 1.47-1.72) and Mediterranean countries (Spain, Portugal, Greece and Italy :average aTFR 1.51 range 1.37-1.65). Mostly, these countries rank slightly lower on the HDI but even when these countries rank high like Switzerland and Spain, they maintain low fertility. Ditto for Japan which has a very HDI, but an extremely conservative family structure.

My point is that cross-country comparison between high-HDI countries are distorted by groups of countries that are culturally similar with similar fertility regimes. The Anglosphere and the Nordic countries have had high and stable fertility -and moreover they had the same level of fertility they do now even at lower levels of development. The presence of these two groups largely explains the correlation between adjusted TFR and HDI. The more conservative countries have had fertility rates that are low and stable. It's therefore difficult to see the impact of HDI on fertility that Myrskylä et al. are claiming

Myrskylä et al. do try to do a longitudinal survey, where they try do look at developments within a country over time and they get similar results. However, the manner in which they conducted the study is, I think, deeply flawed. Since they didn't have a time series of aTFR that went far back enough in time for a sufficient number of countries, they put together a combined series, using TFR when they didn't have adjusted TFR and then controlling for the change in level in the year that they switch between the series. This really seems problematic since TFR and aTFR don't really measure the same thing. Edward pointed this out by looking at the Cohort Fertility rates, but even looking at just the ajusted fertility rate series, it's hard to discern any trend at all. They fact that they got a statistically significant result must be a statistical artifact of the method they used and the limited amount of data available. Just have a look at the graphs for adjusted TFR for the 25 countries in their time series (this is just aTFR and not the combined series they use):

I find it hard to find any support for their argument here. Estonia has experienced an upswing but it happened before the level they claim would induce a fertility increase. The UK series is too short to really tell, but you might be able to make a claim for the US. Either way, the data does not really support their conclusion. I take Jones's point that we can't expect all the hard data we want, but in the absence of it, it would perhaps be better if we all refrained from strong claims without much empirical evidence.


Cicerone said...

Estonia simply rose to fertility-levels normal in other protestant nordic countries. Culturally, Estonia is linked very strongly to Finland, so their demographic regimes are also the same.

Aslak said...

Cicerone, I think that's essentially right. The trough in the beginning of the 90's was presumably caused by the economic trauma of the collapse of the Soviet Union.

J said...

I have a problem with the HDI. What does it measure? Democracy, equality, ecology and other soft, hard to measure social parameters. How come that the HDI has advanced so much in the last 30 years? Why did Sweden, for example, advance twenty or thirty points? Sweden had, in my opinion, a higher HDI thirty years ago than now. A better index should be found, maybe something related to the dissolution of the family framework? We know that the African model of reproduction (informal pairing, short term relations, lack of stability in family members, and so on) produces the largest number of children per woman. Maybe the "africanization" of reproduction is causing the increase? May be that is the natural way we are built.

Aslak said...

J: The HDI really only measure three things: GDP per capita, life expectancy and education as measured by literacy rate and school enrollment rate. As indicators of development go, it's pretty good even if no single indicator will obviously ever capture the whole picture.

Charly said...

There is the problem with HDI that it can´t really be used for temporal comparison.

For instance the problem with education is that it measures literacy and total higher education participation compared to the number of young people. Literacy is something which is highly dependent on the circumstances in somebodies youth. My guess is that the number of Swedish people born in 1900 that stayed illiterate is much higher than those that were born in 1930. Another problem is that they measure total higher education participation compared to total youth population. My guess is that in 1975 the number of 20 year old was higher than the number of 40 year old and my guess is that that is now reversed. This means that now the GEI would be higher even if the participation rate for each age is the same.

For GDP the problem is that it uses GDP. Not that another number would be better cause GDPpp has its own problem

Health has the problem that with time healthcare and hygiene gets better. And it is easier to remove bad housing when population growth slows down which helps with life expectancy

I see especially a problem with GEI as it has a correlation with the tempo effect (quantum effect obviously too)

Mikk said...

Hi, I am estonian, though not a demographer, but I have looked at these things and there is one important thing to mention.

Composition of Estonian population is: 70% estonians and 30% others(mostly russians, but also ukrainians etc.).

These two groups have several differences. For instance russians have lower birth age, more births in marriage and lower TFR. Estonians have higer birth age, more births outside marriage, higher TFR.

If Vienna Demography Institute states that Estonian adjusted TFR is 1,85 and rising (data from 2003-05) then adjusted TFR for estonians alone is even higher and rising.

I have talked with some estonian demographers and they estimate (exact data not available yet) that for estonians alone adjusted TFR is now already 2,0, maybe even higher.
And yes, they confirm that estonian demographic development is more and more similar to other Nordic countries. This similarity was seen already during Soviet era and now even more.