Can predicting data improve model interpretation and inferences? Using Stan to fight the assumption of independence.
We presented an approach which we used to interpret individual effects from models with highly correlated predictors. This poster shows how ignoring a correlation between predictors can induce bias and lead to misleading results. The poster includes links to a ShinyStan application showcasing the models used and to a Shiny application, which allows a specification of the correlation structure between predictors, their association with outcome variable and the effect it has on the model interpretation using the table of coefficients or classical partial dependence plots.
Models used: bit.ly/stan_models
Shiny with simulations: bit.ly/predicting_data