Every study like this needs to adjust for common compounding factors; demographics, income, teaching methods, covid...

Otherwise it's junk data, we have have no idea if lower performance is due to more migrants, more computers, less sugar in cereal or whatever.