Exporting American polarization findings hands other democracies the wrong levers

The weakness the authors identify is bibliographic before it is theoretical. They note that the dominance of US-based or US-focused journals has bent the very definition of political polarization away from European and global contexts, and that many US-focused papers do not explicitly say the US is their subject ^[12]. That is a measurement problem as much as a framing one: a literature you cannot label by country is a literature you cannot easily count. The phys.org account of the work carries no figure for how large the skew is ^[16], so a reader gets a direction without a denominator.

Where the argument does have teeth is the one place a number is not needed, because the sign changed. Deactivating Facebook in the US lowered polarization ^[5]; the same design run in Bosnia-Herzegovina produced the opposite result ^[6][17]. The proposed mechanism is unglamorous and specific: where people have little offline exposure to other ethnicities, the platform was supplying it, and removing the platform removed the contact ^[7]. An offline baseline like that is a variable a US trial holds fixed without ever writing it down.

The thing this does not tell you is where else the flip happens. Two countries establish that direction is contingent, not a map of the contingency, and the material does not show that recommendation algorithms are harmless outside the US. It also does not settle AI's role either way; Falkenberg's claim is narrower, that the polarizing role of AI can only be understood if the evidence is not solely US-focused ^[18], and the authors describe the causal story about algorithms and misinformation as heavily disputed and unlikely to hold in a number of countries ^[15].

Definitions do the rest of the work. Merriam-Webster's 2024 word of the year defined polarization as concentration at two opposing extremes rather than a continuum ^[9], which maps onto a two-party legislature and is an open question in multiparty democracies and power-sharing systems ^[10]. Falkenberg's example is the one-dimensional left-right axis treated as a norm in US work, which he says does not necessarily fit elsewhere ^[4]. The authors also point at Northern Ireland's Good Friday Agreement and separatist disarmament in Spain as episodes that sit awkwardly inside a story of polarization rising everywhere ^[13].

This is an argument about what to fund, not a new dataset. Baronchelli's framing, that institutions, social identities and media environments differ enough that the same intervention can have opposite effects ^[11], leads to a remedy that has a price: promoting and paying for work outside the US ^[14]. A rhetorical correction is free. Panel data in a multiparty system is not.