class: center, middle, inverse, title-slide # Truth be told: Cognitive moderators of selective sharing of fact-checks on social media ### Tiago Ventura ### University of Maryland, College Park ### EGAP - 11/04/2021 --- class: middle, inverse ### When do users share fact-checks with their peers? #### .red[What are the main cognitive moderators of the decision to share corrections?] --- class: middle .pull-left-narrow[ #### .red[Hypothesis] ] .pull-right-wide[ - *H1: Attitude-consistent fact-checks are more likely to be shared than counter-attitudinal fact-checks.* - *H2: Attitude-consistent fact-checks rated ‘true’ are more likely to be shared than attitude-consistent fact-checks rated ‘false.’* ] --- class:middle, center, inverse ## Research Design --- ### Research Design: Experiment + Social Media Data <br><br> Our paper combines two distinct empirical approaches: - .red[On-line Experiment] ~> Two Stage, Two Arm Experiment, where respondents are assigned to True/False Fact-Checking Corrections. - .red[Observational data] ~> Social media from Twitter + a regression discontinuity design. --- ### Experimental Design: General Flow .center[ <img src="figs/flow.png" width="4680" /> ] --- ### Experimental Design: With Images I .pull-left[ .center[ <br> <img src="figs/macri_aud.png" width="80%" /> ] ] .pull-right[ .center[ <img src="figs/true.png" width="50%" /> <img src="figs/false.png" width="50%" /> ] ] --- ### Experimental Design: With Images I .pull-left[ .center[ <br> <img src="figs/ofe.png" width="80%" /> ] ] .pull-right[ .center[ <img src="figs/of_true.png" width="50%" /> <img src="figs/false_of.png" width="50%" /> ] ] --- ### Experimental Design and Hypotheses <br><br> .center[**Would you share?**] <table class="table table-hover" style="font-size: 14px; margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="empty-cells: hide;border-bottom:hidden;" colspan="1"></th> <th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; " colspan="3"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">Audifono</div></th> <th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; " colspan="3"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">Ofelia</div></th> </tr> <tr> <th style="text-align:left;"> Vote Choice </th> <th style="text-align:left;"> Tweet </th> <th style="text-align:left;"> True </th> <th style="text-align:left;"> False </th> <th style="text-align:left;"> Tweet_2 </th> <th style="text-align:left;"> True_Tw2 </th> <th style="text-align:left;"> False_Tw2 </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;width: 1.5in; font-weight: bold;"> Fernández </td> <td style="text-align:left;"> Pro-attitudinal message H1 (+) </td> <td style="text-align:left;"> <span style=" font-weight: bold; color: darkgreen !important;">Pro-attitudinal confirmation<br>H2 (++)</span> </td> <td style="text-align:left;"> Counter-attitudinal refutation </td> <td style="text-align:left;"> Counter-attitudinal message </td> <td style="text-align:left;"> Counter-attitudinal confirmation </td> <td style="text-align:left;"> <span style=" font-weight: bold; color: #C93312 !important;">Pro-attitudinal refutation<br>H2 (+)</span> </td> </tr> <tr> <td style="text-align:left;width: 1.5in; font-weight: bold;"> Macri </td> <td style="text-align:left;"> Counter-attitudinal message </td> <td style="text-align:left;"> Counter-attitudinal confirmation </td> <td style="text-align:left;"> <span style=" font-weight: bold; color: #C93312 !important;">Pro-attitudinal refutation H2 (+)</span> </td> <td style="text-align:left;"> Pro-attitudinal message H1 (+) </td> <td style="text-align:left;"> <span style=" font-weight: bold; color: darkgreen !important;">Pro-attitudinal confirmation<br>H2 (++)</span> </td> <td style="text-align:left;"> Counter-attitudinal refutation </td> </tr> </tbody> </table> --- class:middle, center, inverse ## Results --- ### H1: Pro-Attitudinal Effects .panelset.sideways[ .panel[.panel-name[Audifono] .center[ <img src="./figs/TW1.png" width="80%" /> ] ] .panel[.panel-name[Ofelia] .center[ <img src="./figs/TW2.png" width="100%" /> ] ] ] --- ### H2: Effects of Correction, Macri .panelset.sideways[ .panel[.panel-name[True] .center[ <img src="./figs/true_audifono.png" width="100%" /> ] ] .panel[.panel-name[False] .center[ <img src="./figs/false_audifono.png" width="100%" /> ] ] ] --- ### H2: Effects of Correction, Ofelia .panelset.sideways[ .panel[.panel-name[True] .center[ <img src="./figs/corr_off_false.png" width="100%" /> ] ] .panel[.panel-name[False] .center[ <img src="./figs/corr_off_false_.png" width="100%" /> ] ] ] --- ## Results: Models <img src="./figs/pred_prob.png" width="80%" /> --- ## Results: Latency .center[ <img src="./figs/latency.png" width="65%" /> ] --- class:center, middle, inverse ### Observational Social Media Data --- ### Research Design <br><br> Using the Twitter APIs we collected close to .red[3 million] Tweets during the two weeks around the debate. - Filtered only **retweets** - Searched for the *audifono* misinformation (simple regex search) - Employed **Community Detection algorithms** to identify Pro-Macri and Anti-Macri Using the exact time of the Chequeado Correction (False), we estimated the causal effects (.red[Regression Discontinuity Design]) of the intervention on users' reaction (time-to-retweet) --- ### Results: Time-to-Retweet after the Correction .panelset.sideways[ .panel[.panel-name[Discontinuity Visually] .center[ <img src="./figs/rd_graph.png" width="100%" /> ] ] .panel[.panel-name[Rd Robust Point-Estimates] .center[ <img src="./figs/rd_point.png" width="100%" /> ] ] ] --- ### Conclusion <br><br><br> .panelset.sideways[ .panel[.panel-name[Research Findings] - Users exhibit .red[partisan reasoning] when sharing corrections. - More important, a pro-attitudinal fact-check labeled .red[true] is more likely to be shared on social media than an equally congenial fact-check labeled .red[false] - This result suggests that sharing fact-checking messages is regulated by the .red[hot cognition] and confirmatory bias (directional goals driven by automatic responses) ] .panel[.panel-name[Policy Implications] #### .red[Fact-checkers should consider presenting their work with a ‘true’ adjudication more often] ] ] --- class:inverse, middle, center ## Thank you!