Controlling for exogenous variables is not creepy, but I find the focus on "ethnicity descent" a bit weird when it is the sole parameter being checked. For example, how does this study make sure that there is no other explanatory variable hidden in the country of origin (like e.g. household income) ? He mentions that the US spend a lot of money on education, but since he removes a big part of the students in each country, conclusions are much more difficult to draw: I would expect the money spent per student to be significantly different depending on demographics, for example.
Also, the text uses all encompassing categories with straw man arguments about PISA tests comparison. For example, I can tell you that when the scores went out last year and was discussed in French news, nobody was bragging about how smarter we were compared to the US given how bad the results were for France. He makes some interesting points, but it is hidden within too much authoritative argumentation, which weakens his analysis IMO.
IOW, I think it is an interesting starting point, but it is too crude to make any conclusion, especially the ones he is trying to draw.
Controlling for exogenous variables is not creepy, but I find the focus on "ethnicity descent" a bit weird when it is the sole parameter being checked.
I'm guessing it's less creepy when you take into account that it's just an easily googleable factor.
By the way, the fact that you exclude lots of students is irrelevant, as long as you keep enough to be statistically significant. That's true even when he throws away all non-asian students in the US.
As for household income, that's an even shadier factor to include, since the US is significantly richer than most European nations. Holding income equal, you'd be comparing our 60'th percentile to Sweden's 40'th percentile (numbers made up, but something along those lines).
I don't understand how excluding students is not relevant when discussing about money spent per student: the point is indeed to keep enough to be statistically significant, but w.r.t. what is measured. If money spent by student varies significantly within different demographics, systematically ignoring some demographics alters the validity of the comparison. Maybe it does not, but that needs to be controlled, otherwise it is nonsense statistically speaking.
As for household income, I was not suggesting to use raw numbers, it should of course be normalized, e.g. something like purchase parity, although there may be better ways to do so. It seems difficult to argue that this is shadier than anything related to ethnicity which has no clear definition that I know of.
That is significantly alter the outcome does not make it relevant - it only suggests it may be relevant. The typical example is explaining education success in terms of tv hours views/days during primary education. The variable has a strong prediction power, but it is not really explanatory, as proved when you control with involvement of parents. I feel the same weakness in his analysis.
As for the PPP numbers you gave, they are average, but the whole point of the analysis is to go away from the average, and look at specific demographics. There is no reason to believe they split the same ways in different countries for the same demographic. Maybe the variable is not relevant, but I would be surprised not to see it controlled.
(I would also challenge the fact that US are cheap, but I don't think it is so relevant to the discussion).
Explanatory is a subset of relevant. Furthermore, using a correlated factor as a proxy for the explanatory one is perfectly legitimate if you are attempting to make an unrelated comparison.
(I.e., if "European descent" is correlated with "good home environment" or other exogenous predictors (the data shows it is), and you are comparing EU schools to US schools, it's utterly reasonable to control for "European descent" if data on those exogenous predictors is unavailable.)
As for PPP numbers, the gap often becomes larger when you compare like to like. The same blog did a very good job of this (focusing on not only European Americans, but actually narrowing down to Swedish Americans) a while back:
I don't understand "explanatory is a subset of relevant". If it does not (partially or not) explain the observed result, how can you decide whether it is relevant ?
As for the other links, I don't see numbers related to PPP, but maybe I misunderstand those graphs (I don't understand income per unit of consumption). I also don't see how he can deduce the difference is coming from gvt differences. I am only familiar with statistics, so I may be missing the subtlety of a field I am unfamiliar with (economy), but those analysis seem quite superficial to me. Certainly, they don't warrant such strong conclusions.
Also, the text uses all encompassing categories with straw man arguments about PISA tests comparison. For example, I can tell you that when the scores went out last year and was discussed in French news, nobody was bragging about how smarter we were compared to the US given how bad the results were for France. He makes some interesting points, but it is hidden within too much authoritative argumentation, which weakens his analysis IMO.
IOW, I think it is an interesting starting point, but it is too crude to make any conclusion, especially the ones he is trying to draw.