Gap Minder

Gap Minder is such an awesome tool for viewing cross-country data. Edward Tufte would be proud ([ed]: Is Tufte dead or something? Me: No, I just like saying it that way). Today in class we used it when talking about GDP, and how while Gross Domestic Product is not itself a measure of economic well-being, it is nonetheless highly correlated with and predictive of economic well-being. Well, saying it is one thing, but seeing it is another, and Gap Minder says in a few pictures what I couldn't say in an entire lecture.

Some of the really interesting things in here, though, don't have anything to do with GDP. For instance, plot "fertility rate" on the vertical axis with "child mortality rate" on the horizontal axis, and see the relationship between higher survivorship of children with fertility. I asked students, "Why would a family in a developing country choose to have (for example) 8 children when child mortality is very high, but then actually reduce the number of children they have when mortality rates fall?" I offered as an explanation that there's a certain kind of rationalizing over children in which parents, knowing there's a positive probability their child will die, will produce more children than they want so that in expectation they have the number of children they need. If the odds of death start to fall, then this means they can adjust their optimal fertility risk lower to maintain the same level of children needed. These are, of course, just correlations in the raw data, and I told them they should be cautious in making inference of causality from simple bivariate correlations, but I told them that nonetheless many studies have suggested that this is a causal relationship in which mortality rates actually drive up selected fertility rates.

One other interesting part on here is to plot life expectancy on the vertical axis against time, and look at Rwanda. The genocide has a noticeable drop (understatement). I also did not realize that CO2 was so strongly related to income levels. I knew that many particulate emissions operated by Kuznets curve, wherein increases in income causes increases in pollution up to a point, but after which pollution levels fall with higher levels of income. But apparently this is only for some particulate emissions, because as you can see in Gapminder, CO2 is pretty strongly positively related to income at all levels and for all countries. That is, unless the curve is beyond the level of the US, it does not appear to predict CO2 levels. Of course, I'm not entirely sure what the impact damage of CO2 levels is, but I'm assuming it's something to do with global warming? Anyway, play around with Gap Minder. It's a great resource, and especially helpful as a teaching tool for talking about development and macroeconomics.

Update: Thinking about the problem of correlation and causality and the problem above. So we see higher child mortality being correlated with higher fertility rates, and when mortality falls, so does fertility rates. I then told a story that might explain this - that the lower mortality rates allow parents the option to choose a lower fertility rate because now the "expected" number of children can be chosen with more accuracy, more or less. But, of course, if you plot GNP per person and child mortality, you find more income, lower child mortality. And if you plot GNP per person and fertility, you find (wait for it) higher income, lower fertility. So now I'm asking myself - how much of that correlation between child mortality and total fertility is really just the omitted variable, income? We could find out easily enough using multivariate regression, but not with Gap Minder. Sigh. Such a sweet program nonetheless. Hell, bivariate correlations are valuable, even if all you learn are patterns and relationships. Causality is another matter, and if I wanted to find evidence for causal links between mortality and fertility, then I'd need find a unique shock to mortality that was independent of income. Say, for instance, malaria caused high rates of child mortality, and mosquito nets reduced child mortality. If we could ship nets to developing countries, and observed declines in child mortality (ed: but don't you also need the nets to only affect child mortality, and not adult health, because if the nets also increase adult health, then you're back with the confounding effect of income. Me: Oh shut up you, this is a blog, not a journal. Who are you my conscience now?), then we'd look to see if fertility rates fell too. If not, given some acceptable window of time to observe the change, then we would conclude the mortality-fertility link is not all that strong. But, as my conscience just noted, finding these unique child mortality shocks that are not themselves shocks to adult health is going to be very hard. This is really the challenge with all instrumental variable approaches though. You can move the world if you have a good IV, but unfortunately, they don't really exist. They're like unicorns.