The article linked the study though: https://www.pnas.org/doi/10.1073/pnas.1718793115
So many variables here including way the math portion was measured moving to multiple choice in the 1990s when the scores start dropping, hopping on a flight, still skeptical.
Other quick one, all 95% confidence interval on data that moves less than +/- 2.5% either way. I understand confidence interval isn't a straight linear margin of error, it one I'd want to also look at further.