Brief thread on this "absence of backlash" paper with @andyguess, now out at @BJPolS!
Link to journal: ow.ly/WT5c30mveEs
Ungated: alexandercoppock.com/papers/GC_back…
Link to journal: ow.ly/WT5c30mveEs
Ungated: alexandercoppock.com/papers/GC_back…
1/ We would count *oppositely signed CATE estimates for two different groups* as evidence of backlash to a persuasive attempt. This did not occur in 3 experiments (on gun control, minimum wage, and capital punishment) when we split by pre-treatment measures of policy support.
2/ We also use a fancy Bayesian machine learning approach -- BART -- to try and predict the CATE for every kind of subject in our sample... and we never predict oppositely signed (and significant) treatment effects for any pair of subjects.
3/ We replicate the classic Lord, Ross, and Lepper finding on biased assimilation and attitude polarization (psycnet.apa.org/record/1981-05…). We conclude that study gets exactly the wrong answer because of (a) faulty measurement and (b) lack of random assignment to a control group.
4/ We're not talking about partisan cues, which absolutely do produce oppositely signed effects for Republicans and Democrats -- we're talking about the effects of "persuasive information" that targets a policy attitude.
5/ One fun wrinkle is that people *rate* counter attitudinal evidence very negatively, but are nevertheless persuaded by it, which explains some apparent tensions with previous literature.
6/ We think that *most* treatments will produce "correctly" signed (or null) treatment effects on attitudes *most* of the time! Which means it's probably worth your time to try and persuade the other side -- at least it won't be counterproductive.
