The source he cites is himself, other than the PISA data.
I cannot examine the entire dataset right now, but from reading the questionnaires I cannot find a single question on race. The only resembling question is on the birthplace of the parents [1], but you cannot infer the race from that, can you? How would you separate blacks then, most of whom would probably have American parents?
Edit: The US version of the questionnaire does ask for comprehensive data on race. I still cannot make sense of some of the elements of the Crémieux source, for example they show the US average (not split by race) in a position which is not what I see in the official data. But it should be entirely possible to analyze the data comprehensively by race.
I have known Cremieux for a long time, and trust him a great deal when it comes to handling and presenting data. In any case, I don’t know what standards you’d find satisfactory. He made the graph out of data he listed on the graph. Anyone can go and verity its accuracy. It being materially wrong would be devastating to his reputation. Thus, if you still don’t trust it, you should just do your own leg work to verify it, instead of asking others, whom you probably don’t trust any more than you trust Cremieux, to do it. Anything more would be unusual and unreasonable, even formal academic peer review does not involve verifying that the graphs are accurately representing underlying data.
