Thanks for this reply, but I don’t feel like it really gets at the core of my comment. Looking again at the US data, the different datasets show quite different pictures. Many (correctly, in my opinion) show essentially no recent change; the Polity dataset (screenshotted below) instead shows US democracy as being as low now as in 1798: an era with slavery, no female sufferage, etc.
Even then the different datasets do not agree with your statement though—using the Polity dataset again solely because it’s the one I showed above, we see world levels of democracy at almost their highest levels ever, in contrast to your VDEM claim that democracy has fallen to mid-1980s levels. One of the biggest drivers for the reduction in capita-weighted-democracy for VDEM is India, but Polity actually records India as being extremely democratic—actually more democratic than the US now! Again, this seems clearly absurd to me.
So the mere fact that OWID has a lot of different datasets doesn’t seem very reassuring to me, because 1) the one you specifically highlighted as motivating the project seems problematic, and 2) the others seem to vary significantly and also have significant problems.
I also disagree with your implication that because there are “literally millions of interesting specific data points to look at” that this is some sort of implicit p-hacking. I looked at the US first because it is the most important country, and the problems are apparent on the headline measure; you don’t have to dive into the sub-components. Similarly, I looked at covid because it is clearly one of the most important events of recent years for human freedom, not because I was going around looking for an issue to pick holes over.
Thanks for this reply, but I don’t feel like it really gets at the core of my comment. Looking again at the US data, the different datasets show quite different pictures. Many (correctly, in my opinion) show essentially no recent change; the Polity dataset (screenshotted below) instead shows US democracy as being as low now as in 1798: an era with slavery, no female sufferage, etc.
So the mere fact that OWID has a lot of different datasets doesn’t seem very reassuring to me, because 1) the one you specifically highlighted as motivating the project seems problematic, and 2) the others seem to vary significantly and also have significant problems.
I also disagree with your implication that because there are “literally millions of interesting specific data points to look at” that this is some sort of implicit p-hacking. I looked at the US first because it is the most important country, and the problems are apparent on the headline measure; you don’t have to dive into the sub-components. Similarly, I looked at covid because it is clearly one of the most important events of recent years for human freedom, not because I was going around looking for an issue to pick holes over.