ABSTRACT
This chapter explores three different hazards: multicollinearity, post-treatment bias, and collider bias. It argues that all that is necessary for scientific studies to show a negative association between trustworthiness and newsworthiness is that selection processes—grant and journal review—care about both. The chapter explains how this same selection phenomenon can happen inside a statistical model. In all cases, causal models exist outside the statistical model and can be difficult to test. However, it is possible to reach valid causal inferences in the absence of experiments. The chapter draws on a simple linear regression line through the selected proposals. The selection-distortion effect can happen inside of a multiple regression, because the act of adding a predictor induces statistical selection within the model, a phenomenon that goes by the unhelpful name collider bias.
