The methodology reminded me of Feynman's story about the "Young's 1937 rats study". The conclusions could have been better supported with simpler yet better methodology (given that, of course, from the nature of subject under study it can only reveal correlations, and could fail to detect unknown confounding factors).
They could have actually gone to hundreds of elderly homes in Barcelona or, for that matter, better, in places where everyone is mostly actively bilingual (central/north Catalonia?) and regions where people are mostly monolinguals (Castille?) and requested anonymized data from medical records with that information (actively spoken languages vs. mental condition), or if that information was not available, to actually request it to be collected across all these centers, and then establish much stronger correlations.
edit: I originally wrote "the methodology was unnecessarily weak for the conclusions", which was wrong to say from my side. We cannot always do what would be best but must do with what is available to us.