Huge thank you to collaborators @mmosleh.bsky.social @eckles.bsky.social @dgrand.bsky.social

Comments, feedback, & suggestions appreciated as always!

Caveats:
-Engagement ≠ belief updating (tho it’s an important first step)
-Social corrections can have other negative effects (eg downstream lower quality reposting)
dl.acm.org/doi/abs/10.1...
-Hard to measure (presumably positive) third-party effects of social corrections on observers in field

Our results demonstrate social media’s ability to foster engagement w corrections via minimal social relationships

Ppl are more likely to engage w those who have followed & engaged w them first

A second survey exp found that minimal social connections foster a general norm of responding, such that ppl feel more obligated to respond - and think others expect them to respond more - to ppl who follow them, even outside the context of misinfo correction

Exploratory analyses also show that in both survey & field exps, extreme partisanship moderates the effects of social connection on engagement - social connection increases engagement for co-partisans, but decreases engagement for politically extreme counter-partisans

We next conducted a follow-up survey on MTurk to replicate effects in a more controlled setting (eg eliminate blocking of counter-partisan bots) & obtained similar results

To account for this we (i) compare unaffected conditions (all but social counter-partisan) & (ii) perform principal stratification (weighting obs in unaffected conditions by p(success treat delivery) had they been in social counter-partisan condit)

Users were also more likely to block our bots in the social counter-partisan condition (consistent w our @pnasnexus.org paper on greater blocking of counter-partisans). But this resulted in differential treatment delivery- we could not send corrections to users who blocked our bots shorturl.at/eG5bs

Maybe users simply did not notice or believe the partisanship manip? Prob not: we looked at the follow-back rates in the social condition, & partisanship had a strong effect (consistent w our @pnas.org paper on greater follow-back of copartisans)
shorturl.at/B58Xh

We sent corrections to 1,586 users & measured p(engage w correction):
(i) Among users in the co-partisan condition, social connection had a sig positive effect on engagement
(ii) Among users in the baseline (non-social) condition, no evidence of effect of shared partisanship on engagement

Each user was then socially corrected by their randomly assigned bot. Social corrections were done via public reply to the tweet containing the debunked URL and included a link to the fact-check on @snopes.com

We created human-looking bots & corrected users who shared debunked URLs

We randomized whether our bots
(i) were co-partisan or counter-partisan for the to-be-corrected user
(ii) followed the user & liked some of their tweets before correcting them (creating a minimal social connection)

Social corrections, where users correct one another on social media, have been found to be effective in survey settings shorturl.at/SYomc

But in the field, social corrections are often ignored shorturl.at/jcxPd

We ask what *causes* greater engagement on Twitter (X)

HUGE thank yous to project co-lead @mmosleh.bsky.social (at @oiioxford.bsky.social) & @dgrand.bsky.social

Thoughts, comments, & feedback welcome and appreciated as always!!

Paper here: dx.doi.org/10.1037/xge0...
Preprint here: osf.io/preprints/ps...