What they did
Kahan and colleagues gave 1,111 US adults a two-by-two data table to interpret — a task that requires real numerical reasoning to solve correctly, not a guess. Half the participants were told the data came from a study of a new skin-rash cream; the other half were told the identical numbers came from a study of a gun-control ban's effect on crime. Which conclusion the data actually supported was randomized, so sometimes the "correct" answer favored a more liberal reading and sometimes a more conservative one. Participants were also scored for numeracy — their measured ability and disposition to work with quantitative information.
What they found
- On the neutral skin-cream version, higher numeracy predicted more accurate answers, as expected.
- On the gun-control version, accuracy dropped and split along political lines — but the drop was largest among the most numerate participants, not the least. High-numeracy partisans were about 45% more likely to correctly read the table when the correct answer matched their politics, and worse than low-numeracy partisans when it didn't.
- Their interpretation: more numerate people didn't fail to do the math. They used their stronger math skills selectively — working harder to reach the answer that matched their existing view, and less hard otherwise.
"Subjects highest in Numeracy... did substantially better than less numerate ones when the data were presented as results from a study of a new skin-rash treatment... [but] such polarization did not abate among subjects highest in Numeracy; instead, it increased." — Kahan, Peters, Dawson & Slovic, Behavioural Public Policy 2017
What a later, larger study found
A subsequent large-sample probability study, using the same basic paradigm on a nationally representative US sample, reported no significant association between numerical ability and politically motivated reasoning — a direct non-replication of the original effect's core claim. Both results are published, peer-reviewed work. The honest summary is not "the original was right" or "the original was debunked" — it's that this is a contested, actively-studied finding, and anyone citing it as settled should say so.
Why it's still worth citing carefully
The original finding is widely repeated in commentary about polarization because it's a genuinely striking result — the idea that expertise can sharpen bias rather than correct it runs against a comfortable assumption. That's exactly the kind of claim worth extra scrutiny before repeating, and exactly the kind of claim this page exists to flag when a replication complicates it.