That correlation is not invariant to cultural factors. Your own link mentions some of these - namely, "epistemic belief of learning, performance-avoidance goals, and parental education", for which we can easily measure variation across ethnicity (which of course is a lot more visible and more easily operationalized that a more nuanced understanding of culture, including urban vs. rural).

The paper you linked doesn't report the correlation as far as I could see, but it did cite a paper that said it was 0.86[0]. That's great for a social sciences correlation, but not so great as a way to be fair. 0.86 leaves a lot of room for other processes - and in competitive scenarios at the top of the distribution, the confounders could make all the difference.

[0] https://pubmed.ncbi.nlm.nih.gov/15147489/