Great article. I would suggest correcting the sentence “On the horizontal axis in each panel you see GDP per capita...” as what we see is a logarithm of the GDP per capita. While this doesn’t challenge any of the messages Max is making, one will get a very distorted perception if one assumes that all those metrics are linearly correlated with the GDPpc instead of the logarithm of the GDPpc. In practice, if we consider increasing GDPpc by 1000USD on the left side of the graphs it will shift the point by 30% of the whole range, while if we consider the reduction of the rightmost points by 1000 USD the shift will be ~1%.
This misreading happened to me on the first read and I have also heard other people saying that various metrics are correlated with the GDP, referring to this article.
Another even more nitpicky comment: Although I understand that Max is simply reusing graphics from an external source, I should note that one might be deceived by the presentation in two other cases: - in graphs 2 and 3 as the vertical axis is also logarithmic (implying power dependence on GDP). - in the first and the last two graphs’ Y-axis doesn’t start from zero
Thanks for this important comment! I also agree that it would make the text clearer if we added the fact that we’re dealing with the logarithm of the GDP.
Your observations seem very to the point. Could you elaborate a little bit on what you mean by “implying power dependence on GDP”?
“implying power dependence on GDP” means that the quantity on the Y-axis is the power function of GDP, i.e. GDP^x. It looks like Maternal deaths ~ GDP^(-2), that is for every order of magnitude increase in GDP we have two orders of magnitude decrease in maternal mortality. This is very different from the logarithmic dependence one obtains for straight lines in log-linear plots.
Great article.
I would suggest correcting the sentence “On the horizontal axis in each panel you see GDP per capita...” as what we see is a logarithm of the GDP per capita.
While this doesn’t challenge any of the messages Max is making, one will get a very distorted perception if one assumes that all those metrics are linearly correlated with the GDPpc instead of the logarithm of the GDPpc. In practice, if we consider increasing GDPpc by 1000USD on the left side of the graphs it will shift the point by 30% of the whole range, while if we consider the reduction of the rightmost points by 1000 USD the shift will be ~1%.
This misreading happened to me on the first read and I have also heard other people saying that various metrics are correlated with the GDP, referring to this article.
Another even more nitpicky comment:
Although I understand that Max is simply reusing graphics from an external source, I should note that one might be deceived by the presentation in two other cases:
- in graphs 2 and 3 as the vertical axis is also logarithmic (implying power dependence on GDP).
- in the first and the last two graphs’ Y-axis doesn’t start from zero
Thanks for this important comment! I also agree that it would make the text clearer if we added the fact that we’re dealing with the logarithm of the GDP.
Your observations seem very to the point. Could you elaborate a little bit on what you mean by “implying power dependence on GDP”?
“implying power dependence on GDP” means that the quantity on the Y-axis is the power function of GDP, i.e. GDP^x. It looks like Maternal deaths ~ GDP^(-2), that is for every order of magnitude increase in GDP we have two orders of magnitude decrease in maternal mortality.
This is very different from the logarithmic dependence one obtains for straight lines in log-linear plots.