- [[220427_133707 meta-analysis - warm_repub and ideology|results: warm_repub and ideology for studies 1-4 & study 5]]
# overview of models
- all continuous predictors z-scored within study
- binary variables coded -0.5/0.5
- one bayesian logistic regression multilevel model per study (black)
- `share [0/1] ~ predictors + (1|participant) + (1|stimulus)`
- i.e., cluster on participant and stimulus
- if `veracity [-0.5/0.5]` is a predictor, then `share [0/1] ~ predictors + (1 + veracity|participant) + (1|stimulus)`
- i.e., slope for `veracity` can vary across participants
- meta-analytic multilevel models (red)
- participants nested within studies
- `share [0/1] ~ predictors + (1|study/participant) + (1|stimulus)`
- `share [0/1] ~ predictors + (1+veracity|study/participant) + (1|stimulus)`
- our last [preregistration](https://osf.io/8cnyx)
**Note about `ideology` predictor: In our studies, the `ideology` variable is `demrep_c`. Asher said `warm_repub` is better in studies 1-4 and we included their measure of conservatism in study 5, so we could refit the models using `warm_rep` for studies 1-4 and their measure for study 5.**
# model 1: fail to replicate two-way interaction
- our preregistration model 2 (false and true headlines)
- `share ~ ideology * conscientiousness`
- the simplest effect/model; we can't replicate their basic effect
![[models_demrep-c.png]]
# model 2: don't see three-way-interaction with veracity
- our preregistration model 3 (false and true headlines)
- exclude studies with only false headlines
- `share ~ ideology * conscientiousness * veracity [-0.5/0.5]`
- there's no three-way interaction with veracity
![[models_demrep-c-veracity.png]]
# model 3: three-way interaction with AOT
- exclude studies with only false headlines
- `share ~ ideology * veracity [-0.5/0.5] * aot`
![[models_demrep-veracity-aot.png]]
# model 4: three-way interaction with AOT but not conscientiousness
- exclude studies with only false headlines
- `share ~ ideology * veracity [-0.5/0.5] * (conscientiousness + aot)`
- AOT better predictor than conscientiousness when both predictors are in the model
![[models_full.png]]
# model 5: three-way interaction with AOT but not conscientiousness
- preregistered model 5
- exclude studies with only false headlines
- `share ~ headline veracity [-0.5/0.5] * ideology * (o + c + e + a + n + age + gender + edu + attention + aot)`
- AOT still a better predictor than conscientiousness when both predictors are in the model
- [[220502_102553 probe negative three-way interaction]]
![[models5_full.png]]