The Jupyter notebook in this repo contains a Bayesian mixed-effects re-analysis of the key analysis in Study 3 of Brady, Gantman, & Van Bavel (in press). Data can be obtained from the OSF project.
The main difference between this analysis and the one reported in the paper is the inclusion of random slopes for the 120 normed words. This change is necessary in order to support generalization to a population of words, as opposed to the specific words included in the sample. Including random effects appears to largely eliminate the effect of the norms, suggesting that the reported effects are better explained in terms of between-word variability largely unrelated to attentional capture. Inspection of the posterior distributions confirms that the between-word variance is substantial, with some words (e.g., 'walk', 'killed', and 'moral') showing large positive associations with retweet incidence, and other words (e.g., 'icon', 'alike', and 'theft') showing large negative ones.