Preprint · Jan 8, 2026
Does multilingualism really protect against accelerated ageing? Critical comments on Amoruso et al. (2025)
By Vanhove J
Amoruso et al. (2025; "Multilingualism protects against accelerated aging in cross-sectional and longitudinal analyses of 27 European countries", Nat Aging) purported to have found evidence that multilingualism protects against accelerated ageing. As I will show in this note, several crucial claims made in the article do not stand up to scrutiny, whereas others are easily misunderstood. My criticism consists of the following nine points:1. Missing factors: The article suggests that the accelerated ageing variable was derived from fourteen factors, but the data and code indicate that it is based on only twelve factors.2. Missing respondents and countries: The article claims that the evidence is based on data from 86,149 respondents in 27 European countries. However, the cross-sectional analyses are based on data from only 84,127 respondents in 26 countries; the longitudinal analyses are based on data from only 81,329 respondents.3. Unjustified causal claims: By their own admission in the discussion section, the authors cannot make any causal claims. But the title and abstract contain a bold causal claim.4. Questionable construction of the outcome variable: The outcome variable ("BAGs") is a dichotomised version of a continuous metric. Other aspects of its construction merit scrutiny, too, but will not be dealt with in this note.5. Conflation of country-level and individual multilingualism: The multilingual data that the authors used is situated at the country level, but the authors regularly frame their findings in terms of individual multilingualism.6. Meaningless odds ratios: The odds ratios reported in Amoruso et al.'s Table 1 are presented without any mention of how the data were coded. A closer look at the Jupyter notebooks reveals that they are essentially meaningless.7. Insufficient confounder control: Several confounders are controlled for in the analyses—but only one at a time.8. GDP incorrectly controlled for: The authors claim to have used per capita GDP as a control variable. Their data and code show that they used GDP—not per capita GDP—instead.9. Insufficient accounting for nesting: The nested nature of the data (respondents in cohorts in countries) is not accounted for sufficiently.In the first section, I discuss these points in somewhat more detail; in the second, I rerun a few analyses that Amoruso et al. reported and a few that I consider to be more sensible.