Thread by @BrendanNyhan

Not true - great respect for @archimedino, @sekreps & Kriner but danger of another backfire panic (I would know!)

Paper: dinopc.org/papers/goingtw…

See my annotation - marginal fx of corrections were null for Rs vs control (solid red) & vs seeing unlabeled tweet (dashed red line)

https://twitter.com/WSJopinion/status/1332817609071484929

@JasonReifler & I found backfire effects in 2 of 5 exps in early study fbaum.unc.edu/teaching/artic… Wrongly interpreted to mean backfire common

Further research finds backfire is rare rdcu.be/bhWnf link.springer.com/article/10.100… However still shaping debate/policy eg tweet below

We published an article finding no backfire rdcu.be/bhWnf, I wrote an Upshot column and did an NPR interview on it, and we/I wrote 2 lit reviews em.rdcu.be/wf/click?upn=K… aeaweb.org/articles?id=10… (excerpts below) but no sign of going away. Ironically v. hard to correct!

To be clear, none of that means backfire is impossible! Sometimes people will find it. But in this case, the evidence is not convincing. Important to be clear to avoid furthering the misconception that backfire is common.

More good points here from @JasonReifler

Response 🧵from @archimedino. Two points:
1. Comparison to pure control condition is useful if relevant Q is whether corrections offset damage from misinfo exposure. Marginal evidence there (p=.07) but note the comparison includes *effects of Trump tweet*

https://twitter.com/archimedino/status/1333892630518894592?s=20

2. Comparison to tweet-only condition is proper counterfactual for isolating causal effect of correction. That difference is null. Note: would be worth powering this better! Maybe labels do provoke Rs into believing Trump's claim. Just can't show here.

https://twitter.com/archimedino/status/1333892999856738304?s=20